Johnson111788/Qwen2.5-VL-7B-Instruct-GRPO-OpenImages_3DSR_feb27_60k-2025-02-27-10-40-07
Image-Text-to-Text
•
Updated
•
7
index
stringlengths 21
21
| question
stringlengths 87
406
| A
stringlengths 3
170
| B
stringlengths 2
280
| C
stringclasses 1
value | D
stringclasses 1
value | bounding_box
listlengths 1
3
| direction
listlengths 0
3
| answer
stringclasses 4
values | answer_cot
stringlengths 385
3.54k
| answer_name
stringlengths 6
164
| category
stringclasses 12
values | image_url
stringlengths 20
20
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
00003e2837c7b728_9797 | Consider the real-world 3D locations and orientations of the objects. If I stand at a green stool's position facing where it is facing, is a book with a green spine on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
-0.3,
0.5,
1.1
],
"label": "a book with a green spine"
},
{
"bbox_3d": [
0.9,
0.1,
1.6
],
"label": "a green stool"
}
] | [
{
"front_dir": [
0.1,
0,
-1
],
"label": "a green stool",
"left_dir": [
-1,
0,
-0.1
]
}
] | A | To solve this problem, we first determine the 3D locations of a book with a green spine and a green stool. Then we estimate the vector pointing from a green stool to a book with a green spine, as well as the left direction of a green stool. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a book with a green spine is on the left of a green stool. Otherwise, a book with a green spine is behind a green stool. The 3D location of a book with a green spine is (-0.3, 0.5, 1.1). The 3D location of a green stool is (0.9, 0.1, 1.6). The vector from a green stool to a book with a green spine is hence (-1.1, 0.4, -0.5). The left direction of a green stool is (-1.0, 0.0, -0.1). The cosine similarity between the vector and the left direction is 0.90, corresponding to an angle of 25.87 degrees. The angle is smaller than 90 degrees, meaning that a book with a green spine is on the left of a green stool. Therefore, the final answer is A. on the left. | A. on the left. | orientation_on_the_left | 00003e2837c7b728.jpg |
0000615b5a80f660_2246 | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a metal exhaust hood and a white coffee maker, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
-1.2,
1.7,
3.4
],
"label": "a metal exhaust hood"
},
{
"bbox_3d": [
-1.5,
1.2,
3.2
],
"label": "a white coffee maker"
}
] | [
{
"front_dir": [
0.4,
-0.1,
-0.9
],
"label": "a metal exhaust hood",
"left_dir": [
-0.9,
0.1,
-0.4
]
},
{
"front_dir": [
0.6,
0,
-0.8
],
"label": "a white coffee maker",
"left_dir": [
-0.8,
0.1,
-0.6
]
}
] | A | To solve this problem, we first detect the front directions of a metal exhaust hood and a white coffee maker. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a metal exhaust hood is (0.4, -0.1, -0.9). The front direction of a white coffee maker is (0.6, -0.0, -0.8). The cosine similarity between the two front directions is 0.96, corresponding to an angle of 16.24. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 0000615b5a80f660.jpg |
0000f2101250b009_b7d4 | Consider the real-world 3D locations of the objects. Are the a sheep with a white face and the a herd of sheep in a cage next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
-0.7,
0.6,
1.8
],
"label": "a sheep with a white face"
},
{
"bbox_3d": [
0,
0.6,
1.5
],
"label": "a herd of sheep in a cage"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a sheep with a white face and a herd of sheep in a cage. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a sheep with a white face is (-0.7, 0.6, 1.8). The 3D location of a herd of sheep in a cage is (0.0, 0.6, 1.5). The L2 distance between the two objects is 0.76. The size of the a sheep with a white face is roughly 1.47. The size of the a herd of sheep in a cage is roughly 1.99. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 0000f2101250b009.jpg |
0000f8aef032941e_6e04 | Consider the real-world 3D locations of the objects. Are the a man in a white shirt and blue jeans and the a classroom with a group of people next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
-2.3,
2.5,
6.3
],
"label": "a man in a white shirt and blue jeans"
},
{
"bbox_3d": [
0,
3.4,
8.2
],
"label": "a classroom with a group of people"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a man in a white shirt and blue jeans and a classroom with a group of people. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a man in a white shirt and blue jeans is (-2.3, 2.5, 6.3). The 3D location of a classroom with a group of people is (-0.0, 3.4, 8.2). The L2 distance between the two objects is 3.08. The size of the a man in a white shirt and blue jeans is roughly 1.50. The size of the a classroom with a group of people is roughly 8.50. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 0000f8aef032941e.jpg |
0000f8aef032941e_cf26 | Consider the real-world 3D orientations of the objects. Are a black chair and a black chair with a black seat facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
-1.2,
1.8,
5.4
],
"label": "a black chair"
},
{
"bbox_3d": [
2.3,
1.6,
4.9
],
"label": "a black chair with a black seat"
}
] | [
{
"front_dir": [
0.3,
-0.2,
-0.9
],
"label": "a black chair",
"left_dir": [
-1,
0,
-0.3
]
},
{
"front_dir": [
-0.9,
-0.2,
-0.3
],
"label": "a black chair with a black seat",
"left_dir": [
-0.3,
0.1,
1
]
}
] | B | To solve this problem, we first detect the front directions of a black chair and a black chair with a black seat. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a black chair is (0.3, -0.2, -0.9). The front direction of a black chair with a black seat is (-0.9, -0.2, -0.3). The cosine similarity between the two front directions is 0.04, corresponding to an angle of 87.58. The angle is large, meaning that the two objects are facing very different directions. Therefore, the final answer is B. very different directions. | B. very different directions. | multi_object_same_direction | 0000f8aef032941e.jpg |
00010417d07870a7_387c | Consider the real-world 3D locations of the objects. Are the a man in a green shirt and khaki shorts walking and the a person wearing a green shirt next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
0.3,
1.4,
3
],
"label": "a man in a green shirt and khaki shorts walking"
},
{
"bbox_3d": [
-0.4,
3.1,
19.8
],
"label": "a person wearing a green shirt"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a man in a green shirt and khaki shorts walking and a person wearing a green shirt. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a man in a green shirt and khaki shorts walking is (0.3, 1.4, 3.0). The 3D location of a person wearing a green shirt is (-0.4, 3.1, 19.8). The L2 distance between the two objects is 16.89. The size of the a man in a green shirt and khaki shorts walking is roughly 2.21. The size of the a person wearing a green shirt is roughly 17.90. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 00010417d07870a7.jpg |
00010bf498b64bab_371f | Consider the real-world 3D locations and orientations of the objects. Which side of a black car is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
-1.7,
0.5,
15.3
],
"label": "a black car"
}
] | [
{
"front_dir": [
0.3,
0,
-1
],
"label": "a black car",
"left_dir": [
-1,
0.1,
-0.3
]
}
] | A | To solve this problem, we first estimate the 3D location of a black car. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a black car, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a black car that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a black car is (-1.7, 0.5, 15.3). The vector from a black car to camera is hence (1.7, -0.5, -15.3). The left direction of a black car is (-1.0, 0.1, -0.3). The cosine similarity between the vector pointing to camera and the left direction is 0.16, corresponding to an angle of 80.76 degrees. Thus the angle between the vector pointing to camera and the right direction is 99.24 degrees. The front direction of a black car is (0.3, -0.0, -1.0). The cosine similarity between the vector pointing to camera and the front direction is 0.99, corresponding to an angle of 9.24 degrees. Thus the angle between the vector pointing to camera and the back direction is 170.76 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 9.24 degrees. Thus the front side of a black car is facing the camera. Therefore, the final answer is A. front. | A. front. | orientation_viewpoint | 00010bf498b64bab.jpg |
00046553cda0c8d7_6186 | Consider the real-world 3D locations and orientations of the objects. Which object is a wooden easel with paintings facing towards, a black hat or the a woman wearing a white coat? | a black hat | a woman wearing a white coat | null | null | [
{
"bbox_3d": [
1.2,
0.9,
6.9
],
"label": "a wooden easel with paintings"
},
{
"bbox_3d": [
0,
1.1,
7.5
],
"label": "a black hat"
},
{
"bbox_3d": [
-0.3,
0.5,
4.4
],
"label": "a woman wearing a white coat"
}
] | [
{
"front_dir": [
-0.4,
-0.1,
-0.9
],
"label": "a wooden easel with paintings",
"left_dir": [
-0.9,
0.1,
0.4
]
}
] | B | To solve this problem, we first detect the 3D location of a wooden easel with paintings, a black hat, and a woman wearing a white coat. Then we compute the cosine similarities between the front direction of a wooden easel with paintings and the vectors from a wooden easel with paintings to the other two objects. We can estimate the angles from the cosine similarities, and the object with a smaller angle is the object that a wooden easel with paintings is facing towards. The 3D location of a wooden easel with paintings is (1.2, 0.9, 6.9). The 3D location of a black hat is (0.0, 1.1, 7.5). The 3D location of a woman wearing a white coat is (-0.3, 0.5, 4.4). The front direction of a wooden easel with paintings is (-0.4, -0.1, -0.9). First we consider if a wooden easel with paintings is facing towards the a black hat. The vector from a wooden easel with paintings to a black hat is (-1.2, 0.2, 0.7). The cosine similarity between the front direction and the vector is -0.13, corresponding to an angle of 97.41 degrees. First we consider if a wooden easel with paintings is facing towards the a woman wearing a white coat. The vector from a wooden easel with paintings to a woman wearing a white coat is (-1.5, -0.3, -2.5). The cosine similarity between the front direction and the vector is 0.99, corresponding to an angle of 7.99 degrees. We find that the angle between the front direction and a woman wearing a white coat is smaller. Therefore, the final answer is B. a woman wearing a white coat. | B. a woman wearing a white coat. | multi_object_facing | 00046553cda0c8d7.jpg |
000472189adb5cbd_b785 | Consider the real-world 3D orientations of the objects. Are a black car parked and a red car parked in front of a house facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
9.8,
1.9,
21.4
],
"label": "a black car parked"
},
{
"bbox_3d": [
6.5,
2.2,
24.5
],
"label": "a red car parked in front of a house"
}
] | [
{
"front_dir": [
0.2,
0.1,
1
],
"label": "a black car parked",
"left_dir": [
1,
-0.1,
-0.2
]
},
{
"front_dir": [
0.3,
-0.2,
-0.9
],
"label": "a red car parked in front of a house",
"left_dir": [
-1,
0,
-0.3
]
}
] | B | To solve this problem, we first detect the front directions of a black car parked and a red car parked in front of a house. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a black car parked is (0.2, 0.1, 1.0). The front direction of a red car parked in front of a house is (0.3, -0.2, -0.9). The cosine similarity between the two front directions is -0.89, corresponding to an angle of 152.39. The angle is large, meaning that the two objects are facing very different directions. Therefore, the final answer is B. very different directions. | B. very different directions. | multi_object_same_direction | 000472189adb5cbd.jpg |
00049f0e30c1dd2a_7710 | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a green car and a green car, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
6.4,
4.5,
98.1
],
"label": "a green car"
},
{
"bbox_3d": [
14,
2.4,
38.8
],
"label": "a green car"
}
] | [
{
"front_dir": [
0,
-0.1,
-1
],
"label": "a green car",
"left_dir": [
-1,
0.1,
0
]
},
{
"front_dir": [
-0.3,
-0.1,
-1
],
"label": "a green car",
"left_dir": [
-1,
0.1,
0.3
]
}
] | A | To solve this problem, we first detect the front directions of a green car and a green car. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a green car is (0.0, -0.1, -1.0). The front direction of a green car is (-0.3, -0.1, -1.0). The cosine similarity between the two front directions is 0.96, corresponding to an angle of 16.07. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 00049f0e30c1dd2a.jpg |
00050647c857b018_4d34 | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a woman sitting on a wooden bench and a wooden chair with a woman sitting on it, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
0.3,
0.5,
2.4
],
"label": "a woman sitting on a wooden bench"
},
{
"bbox_3d": [
0,
0.3,
1.4
],
"label": "a wooden chair with a woman sitting on it"
}
] | [
{
"front_dir": [
0,
0,
-1
],
"label": "a woman sitting on a wooden bench",
"left_dir": [
-1,
-0.1,
0
]
},
{
"front_dir": [
0.1,
0.2,
-1
],
"label": "a wooden chair with a woman sitting on it",
"left_dir": [
-1,
-0.1,
-0.1
]
}
] | A | To solve this problem, we first detect the front directions of a woman sitting on a wooden bench and a wooden chair with a woman sitting on it. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a woman sitting on a wooden bench is (-0.0, -0.0, -1.0). The front direction of a wooden chair with a woman sitting on it is (0.1, 0.2, -1.0). The cosine similarity between the two front directions is 0.97, corresponding to an angle of 13.54. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 00050647c857b018.jpg |
0005133a0bf2ecc1_8104 | Consider the real-world 3D locations and orientations of the objects. If I stand at a bulletin board with a picture of a vineyard's position facing where it is facing, is a man and a woman standing together in front of me or behind me? | in front of | behind | null | null | [
{
"bbox_3d": [
-0.2,
0.7,
1.8
],
"label": "a man and a woman standing together"
},
{
"bbox_3d": [
0.5,
0.5,
1.6
],
"label": "a bulletin board with a picture of a vineyard"
}
] | [
{
"front_dir": [
0.7,
0.1,
-0.7
],
"label": "a bulletin board with a picture of a vineyard",
"left_dir": [
-0.8,
0.1,
-0.6
]
}
] | B | To solve this problem, we first determine the 3D locations of a man and a woman standing together and a bulletin board with a picture of a vineyard. Then we estimate the vector pointing from a bulletin board with a picture of a vineyard to a man and a woman standing together, as well as the front direction of a bulletin board with a picture of a vineyard. Next we compute the cosine similarities between the vector and the front direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a man and a woman standing together is in front of a bulletin board with a picture of a vineyard. Otherwise, a man and a woman standing together is behind a bulletin board with a picture of a vineyard. The 3D location of a man and a woman standing together is (-0.2, 0.7, 1.8). The 3D location of a bulletin board with a picture of a vineyard is (0.5, 0.5, 1.6). The vector from a bulletin board with a picture of a vineyard to a man and a woman standing together is hence (-0.8, 0.2, 0.2). The front direction of a bulletin board with a picture of a vineyard is (0.7, 0.1, -0.7). The cosine similarity between the vector and the front direction is -0.77, corresponding to an angle of 140.66 degrees. The angle is smaller than 90 degrees, meaning that a man and a woman standing together is behind a bulletin board with a picture of a vineyard. Therefore, the final answer is B. behind. | B. behind. | orientation_in_front_of | 0005133a0bf2ecc1.jpg |
00056223ec2b5aa1_1be8 | Consider the real-world 3D locations of the objects. Which is closer to a woman in a red shirt sitting in a chair, a seat with a brown leather cover or a train with a large window? | a seat with a brown leather cover | a train with a large window | null | null | [
{
"bbox_3d": [
-0.3,
0.8,
1.9
],
"label": "a woman in a red shirt sitting in a chair"
},
{
"bbox_3d": [
0.6,
0.7,
1.8
],
"label": "a seat with a brown leather cover"
},
{
"bbox_3d": [
0.4,
1.5,
3.5
],
"label": "a train with a large window"
}
] | [] | A | To solve this problem, we first detect the 3D location of a woman in a red shirt sitting in a chair, a seat with a brown leather cover, and a train with a large window. Then we compute the L2 distances between a woman in a red shirt sitting in a chair and a seat with a brown leather cover, and between a woman in a red shirt sitting in a chair and a train with a large window. The object that is closer to a woman in a red shirt sitting in a chair is the one with a smaller distance. The 3D location of a woman in a red shirt sitting in a chair is (-0.3, 0.8, 1.9). The 3D location of a seat with a brown leather cover is (0.6, 0.7, 1.8). The 3D location of a train with a large window is (0.4, 1.5, 3.5). The L2 distance between a woman in a red shirt sitting in a chair and a seat with a brown leather cover is 0.8917982263195952. The L2 distance between a woman in a red shirt sitting in a chair and a train with a large window is 1.8809886209701034. Between the two distances, the distance between a woman in a red shirt sitting in a chair and a seat with a brown leather cover is smaller. Therefore, the final answer is A. a seat with a brown leather cover. | A. a seat with a brown leather cover. | multi_object_closer_to | 00056223ec2b5aa1.jpg |
00056dc4f587f43e_9377 | Consider the real-world 3D locations and orientations of the objects. Which side of a black rectangular object is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
0.4,
1.2,
6.7
],
"label": "a black rectangular object"
}
] | [
{
"front_dir": [
0,
-0.1,
-1
],
"label": "a black rectangular object",
"left_dir": [
-1,
0,
0
]
}
] | A | To solve this problem, we first estimate the 3D location of a black rectangular object. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a black rectangular object, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a black rectangular object that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a black rectangular object is (0.4, 1.2, 6.7). The vector from a black rectangular object to camera is hence (-0.4, -1.2, -6.7). The left direction of a black rectangular object is (-1.0, 0.0, -0.0). The cosine similarity between the vector pointing to camera and the left direction is 0.07, corresponding to an angle of 86.03 degrees. Thus the angle between the vector pointing to camera and the right direction is 93.97 degrees. The front direction of a black rectangular object is (0.0, -0.1, -1.0). The cosine similarity between the vector pointing to camera and the front direction is 0.99, corresponding to an angle of 8.17 degrees. Thus the angle between the vector pointing to camera and the back direction is 171.83 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 8.17 degrees. Thus the front side of a black rectangular object is facing the camera. Therefore, the final answer is A. front. | A. front. | orientation_viewpoint | 00056dc4f587f43e.jpg |
0005dadf2a8e8bef_886a | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a boat with a yellow stripe and a boat with a red hat, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
-0.4,
0.6,
6.4
],
"label": "a boat with a yellow stripe"
},
{
"bbox_3d": [
0.2,
0.5,
6.5
],
"label": "a boat with a red hat"
}
] | [
{
"front_dir": [
0.1,
-0.1,
-1
],
"label": "a boat with a yellow stripe",
"left_dir": [
-1,
0.1,
-0.1
]
},
{
"front_dir": [
0,
-0.1,
-1
],
"label": "a boat with a red hat",
"left_dir": [
-1,
0.1,
0
]
}
] | A | To solve this problem, we first detect the front directions of a boat with a yellow stripe and a boat with a red hat. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a boat with a yellow stripe is (0.1, -0.1, -1.0). The front direction of a boat with a red hat is (0.0, -0.1, -1.0). The cosine similarity between the two front directions is 1.00, corresponding to an angle of 5.27. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 0005dadf2a8e8bef.jpg |
0005fd0beaedf55c_fde0 | Consider the real-world 3D locations of the objects. Are the a beak with a brown texture and the a bird with a pink head next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
0.4,
0.4,
1.7
],
"label": "a beak with a brown texture"
},
{
"bbox_3d": [
0.2,
0.3,
1.5
],
"label": "a bird with a pink head"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a beak with a brown texture and a bird with a pink head. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a beak with a brown texture is (0.4, 0.4, 1.7). The 3D location of a bird with a pink head is (0.2, 0.3, 1.5). The L2 distance between the two objects is 0.25. The size of the a beak with a brown texture is roughly 0.08. The size of the a bird with a pink head is roughly 0.58. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 0005fd0beaedf55c.jpg |
0006567209c36aae_cb14 | Consider the real-world 3D locations of the objects. Are the a man in a white shirt and the a man in a green shirt sitting next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
-0.4,
1.3,
2.9
],
"label": "a man in a white shirt"
},
{
"bbox_3d": [
-0.1,
2.6,
2.3
],
"label": "a man in a green shirt sitting"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a man in a white shirt and a man in a green shirt sitting. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a man in a white shirt is (-0.4, 1.3, 2.9). The 3D location of a man in a green shirt sitting is (-0.1, 2.6, 2.3). The L2 distance between the two objects is 1.41. The size of the a man in a white shirt is roughly 0.67. The size of the a man in a green shirt sitting is roughly 1.23. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 0006567209c36aae.jpg |
0007a2e7a0558219_6741 | Consider the real-world 3D location of the objects. Which object is closer to the camera? | a man in a suit | a bouquet of flowers | null | null | [
{
"bbox_3d": [
-0.3,
1.3,
3.6
],
"label": "a man in a suit"
},
{
"bbox_3d": [
0.1,
0.8,
2.6
],
"label": "a bouquet of flowers"
}
] | [] | B | To solve this problem, we first estimate the 3D locations of a man in a suit and a bouquet of flowers. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a man in a suit is (-0.3, 1.3, 3.6). The 3D location of a bouquet of flowers is (0.1, 0.8, 2.6). The L2 distance from the camera to a man in a suit is 3.83. The L2 distance from the camera to a bouquet of flowers is 2.71. The distance to a bouquet of flowers is smaller. Therefore, the answer is B. a bouquet of flowers. | B. a bouquet of flowers. | location_closer_to_camera | 0007a2e7a0558219.jpg |
0007b54189a67423_2022 | Consider the real-world 3D locations of the objects. Is a blue key on an orange background directly underneath a book with keys on the cover? | yes | no | null | null | [
{
"bbox_3d": [
0,
0.7,
1.1
],
"label": "a book with keys on the cover"
},
{
"bbox_3d": [
0.2,
0.6,
1
],
"label": "a blue key on an orange background"
}
] | [] | B | To solve this problem, we first determine the 3D locations of a book with keys on the cover and a blue key on an orange background. Then we compute the vector pointing from a blue key on an orange background to a book with keys on the cover, as well as the up direction of a blue key on an orange background. We estimate the cosine similarity between the vector and the up direction, which leads to the angle between the two directions. If the angle is small, then a book with keys on the cover is directly above a blue key on an orange background. Otherwise, then a book with keys on the cover is not directly above a blue key on an orange background. To solve the question, we first determine if a book with keys on the cover is directly above a blue key on an orange background. The 3D location of a book with keys on the cover is (0.0, 0.7, 1.1). The 3D location of a blue key on an orange background is (0.2, 0.6, 1.0). The vector from a blue key on an orange background to a book with keys on the cover is hence (-0.2, 0.1, 0.0). The up direction of a blue key on an orange background is (0.0, 1.0, 0.0). The cosine similarity between the vector and the up direction is 0.40, corresponding to an angle of 66 degrees. The angle between the vector and the up direction is large, meaning that a book with keys on the cover is not directly above a blue key on an orange background. In other words, a blue key on an orange background is not directly underneath a book with keys on the cover. Therefore, the answer is B. no. | B. no. | location_above | 0007b54189a67423.jpg |
000820195947013c_f7a0 | Consider the real-world 3D locations of the objects. Are the a baseball player wearing a white uniform and the a baseball player next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
1.8,
1.7,
20
],
"label": "a baseball player wearing a white uniform"
},
{
"bbox_3d": [
1.2,
2.1,
42.5
],
"label": "a baseball player"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a baseball player wearing a white uniform and a baseball player. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a baseball player wearing a white uniform is (1.8, 1.7, 20.0). The 3D location of a baseball player is (1.2, 2.1, 42.5). The L2 distance between the two objects is 22.51. The size of the a baseball player wearing a white uniform is roughly 2.75. The size of the a baseball player is roughly 55.37. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 000820195947013c.jpg |
00084a0cc3a13a3b_f2b0 | Consider the real-world 3D locations of the objects. Are the a woman in a white dress and the a man in a white hat next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
5.2,
1.9,
33.6
],
"label": "a woman in a white dress"
},
{
"bbox_3d": [
5.4,
1.1,
30.7
],
"label": "a man in a white hat"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a woman in a white dress and a man in a white hat. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a woman in a white dress is (5.2, 1.9, 33.6). The 3D location of a man in a white hat is (5.4, 1.1, 30.7). The L2 distance between the two objects is 3.01. The size of the a woman in a white dress is roughly 2.68. The size of the a man in a white hat is roughly 1.06. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 00084a0cc3a13a3b.jpg |
000857e1a9aee1ed_cf43 | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a blue boat with a white bottom and a white boat floating in the water, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
-9.5,
3.8,
53.4
],
"label": "a blue boat with a white bottom"
},
{
"bbox_3d": [
-3.9,
4.1,
49.5
],
"label": "a white boat floating in the water"
}
] | [
{
"front_dir": [
0.2,
-0.1,
-1
],
"label": "a blue boat with a white bottom",
"left_dir": [
-1,
0,
-0.2
]
},
{
"front_dir": [
-1,
0,
-0.3
],
"label": "a white boat floating in the water",
"left_dir": [
-0.3,
0,
1
]
}
] | B | To solve this problem, we first detect the front directions of a blue boat with a white bottom and a white boat floating in the water. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a blue boat with a white bottom is (0.2, -0.1, -1.0). The front direction of a white boat floating in the water is (-1.0, 0.0, -0.3). The cosine similarity between the two front directions is 0.04, corresponding to an angle of 87.49. The angle is closer to 90, than to 0 or 180 degrees, meaning that the two objects are perpendicular to each other. Therefore, the final answer is B. perpendicular. | B. perpendicular. | multi_object_parallel | 000857e1a9aee1ed.jpg |
0008649d2c19d845_68b6 | Consider the real-world 3D locations of the objects. Which object has a higher location? | a white snowy surface | a long wooden plank | null | null | [
{
"bbox_3d": [
0.2,
0.4,
0.6
],
"label": "a white snowy surface"
},
{
"bbox_3d": [
-0.2,
0.5,
0.9
],
"label": "a long wooden plank"
}
] | [] | B | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a white snowy surface is 1.3. The 3D height of a long wooden plank is 1.8. The 3D height of a long wooden plank is larger, meaning that the location of a long wooden plank is higher. Therefore, the answer is B. a long wooden plank. | B. a long wooden plank. | height_higher | 0008649d2c19d845.jpg |
00090aa6938b7de9_ba3f | Consider the real-world 3D locations and orientations of the objects. If I stand at a white comforter on a bed's position facing where it is facing, is a hand with a bandage on it in front of me or behind me? | in front of | behind | null | null | [
{
"bbox_3d": [
-0.2,
0.9,
0.5
],
"label": "a hand with a bandage on it"
},
{
"bbox_3d": [
-0.1,
0.5,
0.9
],
"label": "a white comforter on a bed"
}
] | [
{
"front_dir": [
0.1,
0.7,
-0.7
],
"label": "a white comforter on a bed",
"left_dir": [
-0.9,
0.3,
0.1
]
}
] | A | To solve this problem, we first determine the 3D locations of a hand with a bandage on it and a white comforter on a bed. Then we estimate the vector pointing from a white comforter on a bed to a hand with a bandage on it, as well as the front direction of a white comforter on a bed. Next we compute the cosine similarities between the vector and the front direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a hand with a bandage on it is in front of a white comforter on a bed. Otherwise, a hand with a bandage on it is behind a white comforter on a bed. The 3D location of a hand with a bandage on it is (-0.2, 0.9, 0.5). The 3D location of a white comforter on a bed is (-0.1, 0.5, 0.9). The vector from a white comforter on a bed to a hand with a bandage on it is hence (-0.1, 0.4, -0.3). The front direction of a white comforter on a bed is (0.1, 0.7, -0.7). The cosine similarity between the vector and the front direction is 0.95, corresponding to an angle of 18.18 degrees. The angle is smaller than 90 degrees, meaning that a hand with a bandage on it is in front of a white comforter on a bed. Therefore, the final answer is A. in front of. | A. in front of. | orientation_in_front_of | 00090aa6938b7de9.jpg |
0009c51c7727cce0_ed2d | Consider the real-world 3D locations and orientations of the objects. Which side of a computer with a keyboard is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
-0.1,
0,
0.6
],
"label": "a computer with a keyboard"
}
] | [
{
"front_dir": [
0.2,
0.1,
-1
],
"label": "a computer with a keyboard",
"left_dir": [
-1,
0.1,
-0.2
]
}
] | A | To solve this problem, we first estimate the 3D location of a computer with a keyboard. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a computer with a keyboard, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a computer with a keyboard that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a computer with a keyboard is (-0.1, 0.0, 0.6). The vector from a computer with a keyboard to camera is hence (0.1, -0.0, -0.6). The left direction of a computer with a keyboard is (-1.0, 0.1, -0.2). The cosine similarity between the vector pointing to camera and the left direction is 0.07, corresponding to an angle of 86.24 degrees. Thus the angle between the vector pointing to camera and the right direction is 93.76 degrees. The front direction of a computer with a keyboard is (0.2, 0.1, -1.0). The cosine similarity between the vector pointing to camera and the front direction is 0.99, corresponding to an angle of 8.21 degrees. Thus the angle between the vector pointing to camera and the back direction is 171.79 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 8.21 degrees. Thus the front side of a computer with a keyboard is facing the camera. Therefore, the final answer is A. front. | A. front. | orientation_viewpoint | 0009c51c7727cce0.jpg |
0009d6b2e2f0c698_e6f6 | Consider the real-world 3D locations and orientations of the objects. Which side of a stainless steel exhaust hood is facing a red onion with a white center? | front | left | back | right | [
{
"bbox_3d": [
2.5,
1.8,
7.9
],
"label": "a stainless steel exhaust hood"
},
{
"bbox_3d": [
0.3,
0.7,
1.2
],
"label": "a red onion with a white center"
}
] | [
{
"front_dir": [
-0.2,
-0.3,
-0.9
],
"label": "a stainless steel exhaust hood",
"left_dir": [
-1,
0,
0.2
]
}
] | A | To solve this problem, we first detect the 3D locations of a stainless steel exhaust hood and a red onion with a white center. Then we compute the vector pointing from a stainless steel exhaust hood to a red onion with a white center. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a stainless steel exhaust hood, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a stainless steel exhaust hood that is facing a red onion with a white center corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The 3D location of a stainless steel exhaust hood is (2.5, 1.8, 7.9). The 3D location of a red onion with a white center is (0.3, 0.7, 1.2). The vector from a stainless steel exhaust hood to a red onion with a white center is hence (-2.3, -1.1, -6.7). The left direction of a stainless steel exhaust hood is (-1.0, -0.0, 0.2). The cosine similarity between the vector pointing to a red onion with a white center and the left direction is 0.09, corresponding to an angle of 84.70 degrees. Thus the angle between the vector pointing to a red onion with a white center and the right direction is 95.30 degrees. The front direction of a stainless steel exhaust hood is (-0.2, -0.3, -0.9). The cosine similarity between the vector pointing to a red onion with a white center and the front direction is 0.99, corresponding to an angle of 9.14 degrees. Thus the angle between the vector pointing to a red onion with a white center and the back direction is 170.86 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 9.14 degrees. Thus the front side of a stainless steel exhaust hood is facing the a red onion with a white center. Therefore, the final answer is A. front. | A. front. | multi_object_viewpoint_towards_object | 0009d6b2e2f0c698.jpg |
0009ddf40bd5258a_c8d3 | Consider the real-world 3D location of the objects. Which object is further away from the camera? | a bottle of lotion | a bottle with a blue label | null | null | [
{
"bbox_3d": [
0.9,
0.2,
3.8
],
"label": "a bottle of lotion"
},
{
"bbox_3d": [
-0.6,
1.1,
4.9
],
"label": "a bottle with a blue label"
}
] | [] | B | To solve this problem, we first estimate the 3D locations of a bottle of lotion and a bottle with a blue label. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a bottle of lotion is (0.9, 0.2, 3.8). The 3D location of a bottle with a blue label is (-0.6, 1.1, 4.9). The L2 distance from the camera to a bottle of lotion is 3.93. The L2 distance from the camera to a bottle with a blue label is 5.05. The distance to a bottle with a blue label is larger. Therefore, the answer is B. a bottle with a blue label. | B. a bottle with a blue label. | location_closer_to_camera | 0009ddf40bd5258a.jpg |
000a0945ecb24c23_ed12 | Consider the real-world 3D locations and orientations of the objects. If I stand at a white toilet bowl's position facing where it is facing, is a white toilet with a black seat on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
-0.1,
0.4,
0.2
],
"label": "a white toilet with a black seat"
},
{
"bbox_3d": [
0.4,
0.6,
0.4
],
"label": "a white toilet bowl"
}
] | [
{
"front_dir": [
-0.2,
0.5,
-0.8
],
"label": "a white toilet bowl",
"left_dir": [
-1,
-0.2,
0.2
]
}
] | A | To solve this problem, we first determine the 3D locations of a white toilet with a black seat and a white toilet bowl. Then we estimate the vector pointing from a white toilet bowl to a white toilet with a black seat, as well as the left direction of a white toilet bowl. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a white toilet with a black seat is on the left of a white toilet bowl. Otherwise, a white toilet with a black seat is behind a white toilet bowl. The 3D location of a white toilet with a black seat is (-0.1, 0.4, 0.2). The 3D location of a white toilet bowl is (0.4, 0.6, 0.4). The vector from a white toilet bowl to a white toilet with a black seat is hence (-0.5, -0.2, -0.1). The left direction of a white toilet bowl is (-1.0, -0.2, 0.2). The cosine similarity between the vector and the left direction is 0.92, corresponding to an angle of 23.52 degrees. The angle is smaller than 90 degrees, meaning that a white toilet with a black seat is on the left of a white toilet bowl. Therefore, the final answer is A. on the left. | A. on the left. | orientation_on_the_left | 000a0945ecb24c23.jpg |
000a7b3789b1392e_aaec | Consider the real-world 3D locations of the objects. Which object has a lower location? | a red umbrella | a pink flower with green leaves | null | null | [
{
"bbox_3d": [
1,
5.9,
29
],
"label": "a red umbrella"
},
{
"bbox_3d": [
-10.1,
4.4,
23.7
],
"label": "a pink flower with green leaves"
}
] | [] | A | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a red umbrella is 6.2. The 3D height of a pink flower with green leaves is 14.9. The 3D height of a pink flower with green leaves is larger, meaning that the location of a pink flower with green leaves is higher. In other words, the location of a red umbrella is lower. Therefore, the answer is A. a red umbrella. | A. a red umbrella | height_higher | 000a7b3789b1392e.jpg |
000a9b9566125270_19f6 | Consider the real-world 3D location of the objects. Which object is further away from the camera? | a green plant with a brown bug on it | a gray insect with long antennae | null | null | [
{
"bbox_3d": [
-0.2,
0.4,
1.4
],
"label": "a green plant with a brown bug on it"
},
{
"bbox_3d": [
0.4,
0.4,
1.1
],
"label": "a gray insect with long antennae"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a green plant with a brown bug on it and a gray insect with long antennae. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a green plant with a brown bug on it is (-0.2, 0.4, 1.4). The 3D location of a gray insect with long antennae is (0.4, 0.4, 1.1). The L2 distance from the camera to a green plant with a brown bug on it is 1.48. The L2 distance from the camera to a gray insect with long antennae is 1.18. The distance to a green plant with a brown bug on it is larger. Therefore, the answer is A. a green plant with a brown bug on it. | A. a green plant with a brown bug on it. | location_closer_to_camera | 000a9b9566125270.jpg |
000b05b7f49cdacd_f56b | Consider the real-world 3D locations of the objects. Which object has a higher location? | a rocky cliff with a cave | a man in a black jacket walking | null | null | [
{
"bbox_3d": [
-2.4,
2.8,
27.1
],
"label": "a rocky cliff with a cave"
},
{
"bbox_3d": [
1.8,
0.9,
19.2
],
"label": "a man in a black jacket walking"
}
] | [] | A | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a rocky cliff with a cave is 7.4. The 3D height of a man in a black jacket walking is 2.1. The 3D height of a rocky cliff with a cave is larger, meaning that the location of a rocky cliff with a cave is higher. Therefore, the answer is A. a man in a black jacket walking. | A. a man in a black jacket walking. | height_higher | 000b05b7f49cdacd.jpg |
000b4671075914cd_e216 | Consider the real-world 3D locations and orientations of the objects. If I stand at a computer monitor's position facing where it is facing, is a black microphone with a silver ring around it in front of me or behind me? | in front of | behind | null | null | [
{
"bbox_3d": [
0.1,
0.8,
1.6
],
"label": "a black microphone with a silver ring around it"
},
{
"bbox_3d": [
1.6,
0.7,
4.8
],
"label": "a computer monitor"
}
] | [
{
"front_dir": [
0.2,
-0.2,
-1
],
"label": "a computer monitor",
"left_dir": [
-1,
0,
-0.2
]
}
] | A | To solve this problem, we first determine the 3D locations of a black microphone with a silver ring around it and a computer monitor. Then we estimate the vector pointing from a computer monitor to a black microphone with a silver ring around it, as well as the front direction of a computer monitor. Next we compute the cosine similarities between the vector and the front direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a black microphone with a silver ring around it is in front of a computer monitor. Otherwise, a black microphone with a silver ring around it is behind a computer monitor. The 3D location of a black microphone with a silver ring around it is (0.1, 0.8, 1.6). The 3D location of a computer monitor is (1.6, 0.7, 4.8). The vector from a computer monitor to a black microphone with a silver ring around it is hence (-1.5, 0.1, -3.3). The front direction of a computer monitor is (0.2, -0.2, -1.0). The cosine similarity between the vector and the front direction is 0.77, corresponding to an angle of 39.33 degrees. The angle is smaller than 90 degrees, meaning that a black microphone with a silver ring around it is in front of a computer monitor. Therefore, the final answer is A. in front of. | A. in front of. | orientation_in_front_of | 000b4671075914cd.jpg |
000bb2f7132013dc_adc0 | Consider the real-world 3D locations of the objects. Which object has a lower location? | a tree with white flowers | a hill with a steep slope | null | null | [
{
"bbox_3d": [
0.4,
1.9,
51.3
],
"label": "a tree with white flowers"
},
{
"bbox_3d": [
16.9,
20.8,
173.7
],
"label": "a hill with a steep slope"
}
] | [] | A | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a tree with white flowers is 6.1. The 3D height of a hill with a steep slope is 48.0. The 3D height of a hill with a steep slope is larger, meaning that the location of a hill with a steep slope is higher. In other words, the location of a tree with white flowers is lower. Therefore, the answer is A. a tree with white flowers. | A. a tree with white flowers | height_higher | 000bb2f7132013dc.jpg |
000bcee5bed5446b_f55f | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a black car with a white stripe on the back and a car with a white roof, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
-5.7,
4.6,
91.8
],
"label": "a black car with a white stripe on the back"
},
{
"bbox_3d": [
-4.5,
3.1,
80.9
],
"label": "a car with a white roof"
}
] | [
{
"front_dir": [
0.1,
0,
-1
],
"label": "a black car with a white stripe on the back",
"left_dir": [
-1,
0.1,
-0.1
]
},
{
"front_dir": [
0.1,
-0.1,
-1
],
"label": "a car with a white roof",
"left_dir": [
-1,
0.1,
-0.1
]
}
] | A | To solve this problem, we first detect the front directions of a black car with a white stripe on the back and a car with a white roof. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a black car with a white stripe on the back is (0.1, 0.0, -1.0). The front direction of a car with a white roof is (0.1, -0.1, -1.0). The cosine similarity between the two front directions is 0.99, corresponding to an angle of 7.19. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 000bcee5bed5446b.jpg |
000c5efb8f0d938f_902e | Consider the real-world 3D locations and orientations of the objects. Which side of a bulletin board with a coca cola advertisement is facing a pile of oranges and kiwis? | front | left | back | right | [
{
"bbox_3d": [
0.3,
1,
5.4
],
"label": "a bulletin board with a coca cola advertisement"
},
{
"bbox_3d": [
-0.7,
0.9,
1.6
],
"label": "a pile of oranges and kiwis"
}
] | [
{
"front_dir": [
0,
0,
-1
],
"label": "a bulletin board with a coca cola advertisement",
"left_dir": [
-1,
0.1,
0
]
}
] | A | To solve this problem, we first detect the 3D locations of a bulletin board with a coca cola advertisement and a pile of oranges and kiwis. Then we compute the vector pointing from a bulletin board with a coca cola advertisement to a pile of oranges and kiwis. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a bulletin board with a coca cola advertisement, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a bulletin board with a coca cola advertisement that is facing a pile of oranges and kiwis corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The 3D location of a bulletin board with a coca cola advertisement is (0.3, 1.0, 5.4). The 3D location of a pile of oranges and kiwis is (-0.7, 0.9, 1.6). The vector from a bulletin board with a coca cola advertisement to a pile of oranges and kiwis is hence (-1.0, -0.1, -3.8). The left direction of a bulletin board with a coca cola advertisement is (-1.0, 0.1, -0.0). The cosine similarity between the vector pointing to a pile of oranges and kiwis and the left direction is 0.27, corresponding to an angle of 74.16 degrees. Thus the angle between the vector pointing to a pile of oranges and kiwis and the right direction is 105.84 degrees. The front direction of a bulletin board with a coca cola advertisement is (0.0, -0.0, -1.0). The cosine similarity between the vector pointing to a pile of oranges and kiwis and the front direction is 0.96, corresponding to an angle of 15.85 degrees. Thus the angle between the vector pointing to a pile of oranges and kiwis and the back direction is 164.15 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 15.85 degrees. Thus the front side of a bulletin board with a coca cola advertisement is facing the a pile of oranges and kiwis. Therefore, the final answer is A. front. | A. front. | multi_object_viewpoint_towards_object | 000c5efb8f0d938f.jpg |
000cca22af11a3d9_6b1a | Consider the real-world 3D location of the objects. Which object is further away from the camera? | a black curtain | a wooden floor | null | null | [
{
"bbox_3d": [
-1.4,
2.3,
7.9
],
"label": "a black curtain"
},
{
"bbox_3d": [
0.9,
0.1,
4.2
],
"label": "a wooden floor"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a black curtain and a wooden floor. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a black curtain is (-1.4, 2.3, 7.9). The 3D location of a wooden floor is (0.9, 0.1, 4.2). The L2 distance from the camera to a black curtain is 8.34. The L2 distance from the camera to a wooden floor is 4.30. The distance to a black curtain is larger. Therefore, the answer is A. a black curtain. | A. a black curtain. | location_closer_to_camera | 000cca22af11a3d9.jpg |
000cd0b046e4390b_0311 | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a car on the road and a car on the road, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
0.6,
1.2,
1.2
],
"label": "a car on the road"
},
{
"bbox_3d": [
0.6,
1,
1.1
],
"label": "a car on the road"
}
] | [
{
"front_dir": [
-0.4,
0.1,
-0.9
],
"label": "a car on the road",
"left_dir": [
-0.9,
0,
0.4
]
},
{
"front_dir": [
-0.4,
0,
-0.9
],
"label": "a car on the road",
"left_dir": [
-0.9,
0,
0.4
]
}
] | A | To solve this problem, we first detect the front directions of a car on the road and a car on the road. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a car on the road is (-0.4, 0.1, -0.9). The front direction of a car on the road is (-0.4, -0.0, -0.9). The cosine similarity between the two front directions is 0.99, corresponding to an angle of 8.48. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 000cd0b046e4390b.jpg |
000d608653d67333_1ab0 | Consider the real-world 3D locations and orientations of the objects. Which side of a black chair with a glass table is facing a table with a brown top? | front | left | back | right | [
{
"bbox_3d": [
-0.5,
1,
1.6
],
"label": "a black chair with a glass table"
},
{
"bbox_3d": [
-1.1,
0.6,
4.3
],
"label": "a table with a brown top"
}
] | [
{
"front_dir": [
0.4,
0.3,
-0.9
],
"label": "a black chair with a glass table",
"left_dir": [
-0.9,
0.1,
-0.3
]
}
] | C | To solve this problem, we first detect the 3D locations of a black chair with a glass table and a table with a brown top. Then we compute the vector pointing from a black chair with a glass table to a table with a brown top. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a black chair with a glass table, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a black chair with a glass table that is facing a table with a brown top corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The 3D location of a black chair with a glass table is (-0.5, 1.0, 1.6). The 3D location of a table with a brown top is (-1.1, 0.6, 4.3). The vector from a black chair with a glass table to a table with a brown top is hence (-0.5, -0.4, 2.7). The left direction of a black chair with a glass table is (-0.9, 0.1, -0.3). The cosine similarity between the vector pointing to a table with a brown top and the left direction is -0.17, corresponding to an angle of 100.05 degrees. Thus the angle between the vector pointing to a table with a brown top and the right direction is 79.95 degrees. The front direction of a black chair with a glass table is (0.4, 0.3, -0.9). The cosine similarity between the vector pointing to a table with a brown top and the front direction is -0.97, corresponding to an angle of 166.48 degrees. Thus the angle between the vector pointing to a table with a brown top and the back direction is 13.52 degrees. Among the four directions, the smallest angle is the back direction, with an angle of 13.52 degrees. Thus the back side of a black chair with a glass table is facing the a table with a brown top. Therefore, the final answer is C. back. | C. back. | multi_object_viewpoint_towards_object | 000d608653d67333.jpg |
000e082b0dce859e_edfe | Consider the real-world 3D locations and orientations of the objects. If I stand at a wooden stool with a metal brace's position facing where it is facing, is a wooden table with a cutting board on it in front of me or behind me? | in front of | behind | null | null | [
{
"bbox_3d": [
-1.2,
0.9,
3.3
],
"label": "a wooden table with a cutting board on it"
},
{
"bbox_3d": [
0,
0.3,
1.3
],
"label": "a wooden stool with a metal brace"
}
] | [
{
"front_dir": [
0.8,
0.1,
-0.5
],
"label": "a wooden stool with a metal brace",
"left_dir": [
-0.6,
0.2,
-0.8
]
}
] | B | To solve this problem, we first determine the 3D locations of a wooden table with a cutting board on it and a wooden stool with a metal brace. Then we estimate the vector pointing from a wooden stool with a metal brace to a wooden table with a cutting board on it, as well as the front direction of a wooden stool with a metal brace. Next we compute the cosine similarities between the vector and the front direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a wooden table with a cutting board on it is in front of a wooden stool with a metal brace. Otherwise, a wooden table with a cutting board on it is behind a wooden stool with a metal brace. The 3D location of a wooden table with a cutting board on it is (-1.2, 0.9, 3.3). The 3D location of a wooden stool with a metal brace is (0.0, 0.3, 1.3). The vector from a wooden stool with a metal brace to a wooden table with a cutting board on it is hence (-1.2, 0.7, 2.0). The front direction of a wooden stool with a metal brace is (0.8, 0.1, -0.5). The cosine similarity between the vector and the front direction is -0.84, corresponding to an angle of 147.12 degrees. The angle is smaller than 90 degrees, meaning that a wooden table with a cutting board on it is behind a wooden stool with a metal brace. Therefore, the final answer is B. behind. | B. behind. | orientation_in_front_of | 000e082b0dce859e.jpg |
000e70d6b9f25c3d_815e | Consider the real-world 3D orientations of the objects. Are a white couch with a brown pillow and a fireplace with a black metal grate and a brick wall facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
-1.2,
0.5,
4.1
],
"label": "a white couch with a brown pillow"
},
{
"bbox_3d": [
0.2,
0.6,
6.3
],
"label": "a fireplace with a black metal grate and a brick wall"
}
] | [
{
"front_dir": [
0.2,
0,
-1
],
"label": "a white couch with a brown pillow",
"left_dir": [
-1,
-0.1,
-0.2
]
},
{
"front_dir": [
-0.1,
0.1,
-1
],
"label": "a fireplace with a black metal grate and a brick wall",
"left_dir": [
-1,
0,
0.1
]
}
] | A | To solve this problem, we first detect the front directions of a white couch with a brown pillow and a fireplace with a black metal grate and a brick wall. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a white couch with a brown pillow is (0.2, 0.0, -1.0). The front direction of a fireplace with a black metal grate and a brick wall is (-0.1, 0.1, -1.0). The cosine similarity between the two front directions is 0.94, corresponding to an angle of 19.79. The angle is small, meaning that the two objects are facing same or similar directions. Therefore, the final answer is A. same or similar directions. | A. same or similar directions. | multi_object_same_direction | 000e70d6b9f25c3d.jpg |
000e842c55ab7d14_5e6c | Consider the real-world 3D locations and orientations of the objects. Which side of a red sports car is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
3.3,
0.8,
5.7
],
"label": "a red sports car"
}
] | [
{
"front_dir": [
0.9,
0.1,
-0.4
],
"label": "a red sports car",
"left_dir": [
-0.4,
-0.1,
-0.9
]
}
] | B | To solve this problem, we first estimate the 3D location of a red sports car. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a red sports car, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a red sports car that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a red sports car is (3.3, 0.8, 5.7). The vector from a red sports car to camera is hence (-3.3, -0.8, -5.7). The left direction of a red sports car is (-0.4, -0.1, -0.9). The cosine similarity between the vector pointing to camera and the left direction is 1.00, corresponding to an angle of 4.59 degrees. Thus the angle between the vector pointing to camera and the right direction is 175.41 degrees. The front direction of a red sports car is (0.9, 0.1, -0.4). The cosine similarity between the vector pointing to camera and the front direction is -0.08, corresponding to an angle of 94.58 degrees. Thus the angle between the vector pointing to camera and the back direction is 85.42 degrees. Among the four directions, the smallest angle is the left direction, with an angle of 4.59 degrees. Thus the left side of a red sports car is facing the camera. Therefore, the final answer is B. left. | B. left. | orientation_viewpoint | 000e842c55ab7d14.jpg |
000edafa589fc672_8c57 | Consider the real-world 3D locations and orientations of the objects. Which side of a metal chair with a wooden seat is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
4.4,
0.7,
8.9
],
"label": "a metal chair with a wooden seat"
}
] | [
{
"front_dir": [
0.9,
0,
-0.5
],
"label": "a metal chair with a wooden seat",
"left_dir": [
-0.5,
0.2,
-0.8
]
}
] | B | To solve this problem, we first estimate the 3D location of a metal chair with a wooden seat. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a metal chair with a wooden seat, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a metal chair with a wooden seat that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a metal chair with a wooden seat is (4.4, 0.7, 8.9). The vector from a metal chair with a wooden seat to camera is hence (-4.4, -0.7, -8.9). The left direction of a metal chair with a wooden seat is (-0.5, 0.2, -0.8). The cosine similarity between the vector pointing to camera and the left direction is 0.95, corresponding to an angle of 18.64 degrees. Thus the angle between the vector pointing to camera and the right direction is 161.36 degrees. The front direction of a metal chair with a wooden seat is (0.9, 0.0, -0.5). The cosine similarity between the vector pointing to camera and the front direction is 0.07, corresponding to an angle of 85.73 degrees. Thus the angle between the vector pointing to camera and the back direction is 94.27 degrees. Among the four directions, the smallest angle is the left direction, with an angle of 18.64 degrees. Thus the left side of a metal chair with a wooden seat is facing the camera. Therefore, the final answer is B. left. | B. left. | orientation_viewpoint | 000edafa589fc672.jpg |
000edafa589fc672_993f | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a chair with a white seat and a red chair with a metal frame, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
1.7,
1.3,
3.4
],
"label": "a chair with a white seat"
},
{
"bbox_3d": [
0.2,
0.5,
8
],
"label": "a red chair with a metal frame"
}
] | [
{
"front_dir": [
-0.8,
0.2,
-0.5
],
"label": "a chair with a white seat",
"left_dir": [
-0.5,
-0.1,
0.8
]
},
{
"front_dir": [
0,
-0.3,
-0.9
],
"label": "a red chair with a metal frame",
"left_dir": [
-1,
0.1,
0
]
}
] | B | To solve this problem, we first detect the front directions of a chair with a white seat and a red chair with a metal frame. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a chair with a white seat is (-0.8, 0.2, -0.5). The front direction of a red chair with a metal frame is (0.0, -0.3, -0.9). The cosine similarity between the two front directions is 0.43, corresponding to an angle of 64.47. The angle is closer to 90, than to 0 or 180 degrees, meaning that the two objects are perpendicular to each other. Therefore, the final answer is B. perpendicular. | B. perpendicular. | multi_object_parallel | 000edafa589fc672.jpg |
000edafa589fc672_0e8a | Consider the real-world 3D orientations of the objects. Are a metal chair with a red seat and a red chair with a white jacket draped over it facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
0.8,
0.2,
9.2
],
"label": "a metal chair with a red seat"
},
{
"bbox_3d": [
1.1,
1,
2.9
],
"label": "a red chair with a white jacket draped over it"
}
] | [
{
"front_dir": [
-1,
0.1,
0
],
"label": "a metal chair with a red seat",
"left_dir": [
0,
0,
1
]
},
{
"front_dir": [
-0.3,
0,
-1
],
"label": "a red chair with a white jacket draped over it",
"left_dir": [
-1,
0,
0.3
]
}
] | B | To solve this problem, we first detect the front directions of a metal chair with a red seat and a red chair with a white jacket draped over it. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a metal chair with a red seat is (-1.0, 0.1, 0.0). The front direction of a red chair with a white jacket draped over it is (-0.3, 0.0, -1.0). The cosine similarity between the two front directions is 0.25, corresponding to an angle of 75.70. The angle is large, meaning that the two objects are facing very different directions. Therefore, the final answer is B. very different directions. | B. very different directions. | multi_object_same_direction | 000edafa589fc672.jpg |
000f63dc614130ae_e16a | Consider the real-world 3D locations and orientations of the objects. Which side of a chair with a man sitting in it is facing a brown chair? | front | left | back | right | [
{
"bbox_3d": [
-0.4,
0.6,
1.6
],
"label": "a chair with a man sitting in it"
},
{
"bbox_3d": [
1.1,
0.4,
1.9
],
"label": "a brown chair"
}
] | [
{
"front_dir": [
1,
-0.2,
0.2
],
"label": "a chair with a man sitting in it",
"left_dir": [
0,
-0.6,
-0.8
]
},
{
"front_dir": [
0,
0.1,
-1
],
"label": "a brown chair",
"left_dir": [
-1,
-0.1,
-0.1
]
}
] | A | To solve this problem, we first detect the 3D locations of a chair with a man sitting in it and a brown chair. Then we compute the vector pointing from a chair with a man sitting in it to a brown chair. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a chair with a man sitting in it, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a chair with a man sitting in it that is facing a brown chair corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The 3D location of a chair with a man sitting in it is (-0.4, 0.6, 1.6). The 3D location of a brown chair is (1.1, 0.4, 1.9). The vector from a chair with a man sitting in it to a brown chair is hence (1.4, -0.3, 0.3). The left direction of a chair with a man sitting in it is (-0.0, -0.6, -0.8). The cosine similarity between the vector pointing to a brown chair and the left direction is -0.06, corresponding to an angle of 93.60 degrees. Thus the angle between the vector pointing to a brown chair and the right direction is 86.40 degrees. The front direction of a chair with a man sitting in it is (1.0, -0.2, 0.2). The cosine similarity between the vector pointing to a brown chair and the front direction is 1.00, corresponding to an angle of 3.97 degrees. Thus the angle between the vector pointing to a brown chair and the back direction is 176.03 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 3.97 degrees. Thus the front side of a chair with a man sitting in it is facing the a brown chair. Therefore, the final answer is A. front. | A. front. | multi_object_viewpoint_towards_object | 000f63dc614130ae.jpg |
000f96aacadf7aa5_14cd | Consider the real-world 3D location of the objects. Which object is closer to the camera? | a white jeep with luggage on top | a dirt field with a tractor | null | null | [
{
"bbox_3d": [
-3.5,
2.4,
10.7
],
"label": "a white jeep with luggage on top"
},
{
"bbox_3d": [
-3.2,
0.1,
5.8
],
"label": "a dirt field with a tractor"
}
] | [] | B | To solve this problem, we first estimate the 3D locations of a white jeep with luggage on top and a dirt field with a tractor. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a white jeep with luggage on top is (-3.5, 2.4, 10.7). The 3D location of a dirt field with a tractor is (-3.2, 0.1, 5.8). The L2 distance from the camera to a white jeep with luggage on top is 11.54. The L2 distance from the camera to a dirt field with a tractor is 6.68. The distance to a dirt field with a tractor is smaller. Therefore, the answer is B. a dirt field with a tractor. | B. a dirt field with a tractor. | location_closer_to_camera | 000f96aacadf7aa5.jpg |
000fbfa6ec5b6457_dc19 | Consider the real-world 3D locations of the objects. Are the a green bottle with a white label and the a woman in a black shirt looking at her phone next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
0,
0.9,
1.5
],
"label": "a green bottle with a white label"
},
{
"bbox_3d": [
0.1,
1,
2.8
],
"label": "a woman in a black shirt looking at her phone"
}
] | [] | B | To solve this problem, we first estimate the 3D locations of a green bottle with a white label and a woman in a black shirt looking at her phone. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a green bottle with a white label is (-0.0, 0.9, 1.5). The 3D location of a woman in a black shirt looking at her phone is (0.1, 1.0, 2.8). The L2 distance between the two objects is 1.39. The size of the a green bottle with a white label is roughly 0.30. The size of the a woman in a black shirt looking at her phone is roughly 0.80. The distance between the two objects is much bigger than the object sizes, meaning that they are far away from each other. Therefore, the answer is B. far away from each other. | B. far away from each other. | location_next_to | 000fbfa6ec5b6457.jpg |
000fd820b7972143_6699 | Consider the real-world 3D locations of the objects. Is a man in a black suit directly underneath a hand with a red tie? | yes | no | null | null | [
{
"bbox_3d": [
0.3,
0.8,
2.1
],
"label": "a hand with a red tie"
},
{
"bbox_3d": [
0.3,
0.6,
2.2
],
"label": "a man in a black suit"
}
] | [] | B | To solve this problem, we first determine the 3D locations of a hand with a red tie and a man in a black suit. Then we compute the vector pointing from a man in a black suit to a hand with a red tie, as well as the up direction of a man in a black suit. We estimate the cosine similarity between the vector and the up direction, which leads to the angle between the two directions. If the angle is small, then a hand with a red tie is directly above a man in a black suit. Otherwise, then a hand with a red tie is not directly above a man in a black suit. To solve the question, we first determine if a hand with a red tie is directly above a man in a black suit. The 3D location of a hand with a red tie is (0.3, 0.8, 2.1). The 3D location of a man in a black suit is (0.3, 0.6, 2.2). The vector from a man in a black suit to a hand with a red tie is hence (-0.0, 0.2, -0.1). The up direction of a man in a black suit is (0.0, 1.0, 0.0). The cosine similarity between the vector and the up direction is 0.73, corresponding to an angle of 43 degrees. The angle between the vector and the up direction is large, meaning that a hand with a red tie is not directly above a man in a black suit. In other words, a man in a black suit is not directly underneath a hand with a red tie. Therefore, the answer is B. no. | B. no. | location_above | 000fd820b7972143.jpg |
000fe11025f2e246_6133 | Consider the real-world 3D locations and orientations of the objects. Which object is a dog with a long tail facing towards, a woman wearing a green shirt or the a busy street with people riding motorcycles? | a woman wearing a green shirt | a busy street with people riding motorcycles | null | null | [
{
"bbox_3d": [
-6.4,
1,
30.6
],
"label": "a dog with a long tail"
},
{
"bbox_3d": [
0.1,
1.1,
7.3
],
"label": "a woman wearing a green shirt"
},
{
"bbox_3d": [
0.3,
5.9,
31.4
],
"label": "a busy street with people riding motorcycles"
}
] | [
{
"front_dir": [
0.3,
0,
-1
],
"label": "a dog with a long tail",
"left_dir": [
-1,
0,
-0.3
]
}
] | A | To solve this problem, we first detect the 3D location of a dog with a long tail, a woman wearing a green shirt, and a busy street with people riding motorcycles. Then we compute the cosine similarities between the front direction of a dog with a long tail and the vectors from a dog with a long tail to the other two objects. We can estimate the angles from the cosine similarities, and the object with a smaller angle is the object that a dog with a long tail is facing towards. The 3D location of a dog with a long tail is (-6.4, 1.0, 30.6). The 3D location of a woman wearing a green shirt is (0.1, 1.1, 7.3). The 3D location of a busy street with people riding motorcycles is (0.3, 5.9, 31.4). The front direction of a dog with a long tail is (0.3, -0.0, -1.0). First we consider if a dog with a long tail is facing towards the a woman wearing a green shirt. The vector from a dog with a long tail to a woman wearing a green shirt is (6.5, 0.2, -23.3). The cosine similarity between the front direction and the vector is 1.00, corresponding to an angle of 2.17 degrees. First we consider if a dog with a long tail is facing towards the a busy street with people riding motorcycles. The vector from a dog with a long tail to a busy street with people riding motorcycles is (6.7, 5.0, 0.8). The cosine similarity between the front direction and the vector is 0.11, corresponding to an angle of 83.41 degrees. We find that the angle between the front direction and a woman wearing a green shirt is smaller. Therefore, the final answer is A. a woman wearing a green shirt. | A. a woman wearing a green shirt. | multi_object_facing | 000fe11025f2e246.jpg |
000fe628fd1fa9d2_e977 | Consider the real-world 3D locations and orientations of the objects. Which side of a white van is driving on the road is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
-0.4,
2.1,
16.3
],
"label": "a white van is driving on the road"
}
] | [
{
"front_dir": [
-0.1,
0.2,
1
],
"label": "a white van is driving on the road",
"left_dir": [
1,
-0.1,
0.1
]
}
] | C | To solve this problem, we first estimate the 3D location of a white van is driving on the road. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a white van is driving on the road, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a white van is driving on the road that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a white van is driving on the road is (-0.4, 2.1, 16.3). The vector from a white van is driving on the road to camera is hence (0.4, -2.1, -16.3). The left direction of a white van is driving on the road is (1.0, -0.1, 0.1). The cosine similarity between the vector pointing to camera and the left direction is -0.08, corresponding to an angle of 94.41 degrees. Thus the angle between the vector pointing to camera and the right direction is 85.59 degrees. The front direction of a white van is driving on the road is (-0.1, 0.2, 1.0). The cosine similarity between the vector pointing to camera and the front direction is -1.00, corresponding to an angle of 174.77 degrees. Thus the angle between the vector pointing to camera and the back direction is 5.23 degrees. Among the four directions, the smallest angle is the back direction, with an angle of 5.23 degrees. Thus the back side of a white van is driving on the road is facing the camera. Therefore, the final answer is C. back. | C. back. | orientation_viewpoint | 000fe628fd1fa9d2.jpg |
001021cac2f74637_b685 | Consider the real-world 3D locations and orientations of the objects. If I stand at a stool with a brown wooden top's position facing where it is facing, is a white sock with a yellow stripe on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
0.3,
-0.1,
1.3
],
"label": "a white sock with a yellow stripe"
},
{
"bbox_3d": [
-0.5,
-0.1,
2
],
"label": "a stool with a brown wooden top"
}
] | [
{
"front_dir": [
0,
0.8,
0.6
],
"label": "a stool with a brown wooden top",
"left_dir": [
0.3,
0.6,
-0.8
]
}
] | A | To solve this problem, we first determine the 3D locations of a white sock with a yellow stripe and a stool with a brown wooden top. Then we estimate the vector pointing from a stool with a brown wooden top to a white sock with a yellow stripe, as well as the left direction of a stool with a brown wooden top. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a white sock with a yellow stripe is on the left of a stool with a brown wooden top. Otherwise, a white sock with a yellow stripe is behind a stool with a brown wooden top. The 3D location of a white sock with a yellow stripe is (0.3, -0.1, 1.3). The 3D location of a stool with a brown wooden top is (-0.5, -0.1, 2.0). The vector from a stool with a brown wooden top to a white sock with a yellow stripe is hence (0.8, 0.0, -0.7). The left direction of a stool with a brown wooden top is (0.3, 0.6, -0.8). The cosine similarity between the vector and the left direction is 0.72, corresponding to an angle of 44.03 degrees. The angle is smaller than 90 degrees, meaning that a white sock with a yellow stripe is on the left of a stool with a brown wooden top. Therefore, the final answer is A. on the left. | A. on the left. | orientation_on_the_left | 001021cac2f74637.jpg |
0010492df4e1b810_c9d8 | Consider the real-world 3D locations and orientations of the objects. Which object is a blue stool facing towards, a black tablecloth on a table or the a black camera with a yellow cord? | a black tablecloth on a table | a black camera with a yellow cord | null | null | [
{
"bbox_3d": [
0.3,
1,
2.4
],
"label": "a blue stool"
},
{
"bbox_3d": [
3.6,
0.8,
6.6
],
"label": "a black tablecloth on a table"
},
{
"bbox_3d": [
0.9,
1.1,
2.1
],
"label": "a black camera with a yellow cord"
}
] | [
{
"front_dir": [
1,
0,
-0.2
],
"label": "a blue stool",
"left_dir": [
-0.2,
0.1,
-1
]
},
{
"front_dir": [
1,
0,
-0.3
],
"label": "a black camera with a yellow cord",
"left_dir": [
-0.3,
0.3,
-0.9
]
}
] | B | To solve this problem, we first detect the 3D location of a blue stool, a black tablecloth on a table, and a black camera with a yellow cord. Then we compute the cosine similarities between the front direction of a blue stool and the vectors from a blue stool to the other two objects. We can estimate the angles from the cosine similarities, and the object with a smaller angle is the object that a blue stool is facing towards. The 3D location of a blue stool is (0.3, 1.0, 2.4). The 3D location of a black tablecloth on a table is (3.6, 0.8, 6.6). The 3D location of a black camera with a yellow cord is (0.9, 1.1, 2.1). The front direction of a blue stool is (1.0, -0.0, -0.2). First we consider if a blue stool is facing towards the a black tablecloth on a table. The vector from a blue stool to a black tablecloth on a table is (3.3, -0.2, 4.1). The cosine similarity between the front direction and the vector is 0.44, corresponding to an angle of 63.83 degrees. First we consider if a blue stool is facing towards the a black camera with a yellow cord. The vector from a blue stool to a black camera with a yellow cord is (0.6, 0.1, -0.3). The cosine similarity between the front direction and the vector is 0.96, corresponding to an angle of 16.29 degrees. We find that the angle between the front direction and a black camera with a yellow cord is smaller. Therefore, the final answer is B. a black camera with a yellow cord. | B. a black camera with a yellow cord. | multi_object_facing | 0010492df4e1b810.jpg |
0010f4c10f7ab07e_04d4 | Consider the real-world 3D orientations of the objects. Are a silver car parked on the street and a brown car facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
3,
0.9,
6.2
],
"label": "a silver car parked on the street"
},
{
"bbox_3d": [
12.4,
0.4,
32.3
],
"label": "a brown car"
}
] | [
{
"front_dir": [
-0.4,
0,
-0.9
],
"label": "a silver car parked on the street",
"left_dir": [
-0.9,
0,
0.4
]
},
{
"front_dir": [
0,
-0.1,
-1
],
"label": "a brown car",
"left_dir": [
-1,
0,
0
]
}
] | A | To solve this problem, we first detect the front directions of a silver car parked on the street and a brown car. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a silver car parked on the street is (-0.4, 0.0, -0.9). The front direction of a brown car is (0.0, -0.1, -1.0). The cosine similarity between the two front directions is 0.92, corresponding to an angle of 23.28. The angle is small, meaning that the two objects are facing same or similar directions. Therefore, the final answer is A. same or similar directions. | A. same or similar directions. | multi_object_same_direction | 0010f4c10f7ab07e.jpg |
00119d989d1e515e_01e9 | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a red car with a black interior and a silver car, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
1.6,
1.2,
6.5
],
"label": "a red car with a black interior"
},
{
"bbox_3d": [
0.9,
1.7,
7
],
"label": "a silver car"
}
] | [
{
"front_dir": [
-0.2,
0,
-1
],
"label": "a red car with a black interior",
"left_dir": [
-1,
0.1,
0.2
]
},
{
"front_dir": [
-0.1,
-0.1,
-1
],
"label": "a silver car",
"left_dir": [
-1,
0.1,
0
]
}
] | A | To solve this problem, we first detect the front directions of a red car with a black interior and a silver car. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a red car with a black interior is (-0.2, -0.0, -1.0). The front direction of a silver car is (-0.1, -0.1, -1.0). The cosine similarity between the two front directions is 0.99, corresponding to an angle of 7.93. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 00119d989d1e515e.jpg |
0011ace1c64f7919_4f8f | Consider the real-world 3D locations of the objects. Which is closer to a cartoon of a goose, a canopy with lights on it or a calm lake with a few trees? | a canopy with lights on it | a calm lake with a few trees | null | null | [
{
"bbox_3d": [
-9.4,
1.1,
23.2
],
"label": "a cartoon of a goose"
},
{
"bbox_3d": [
1.4,
2.1,
6.5
],
"label": "a canopy with lights on it"
},
{
"bbox_3d": [
-8.4,
3.3,
58
],
"label": "a calm lake with a few trees"
}
] | [] | A | To solve this problem, we first detect the 3D location of a cartoon of a goose, a canopy with lights on it, and a calm lake with a few trees. Then we compute the L2 distances between a cartoon of a goose and a canopy with lights on it, and between a cartoon of a goose and a calm lake with a few trees. The object that is closer to a cartoon of a goose is the one with a smaller distance. The 3D location of a cartoon of a goose is (-9.4, 1.1, 23.2). The 3D location of a canopy with lights on it is (1.4, 2.1, 6.5). The 3D location of a calm lake with a few trees is (-8.4, 3.3, 58.0). The L2 distance between a cartoon of a goose and a canopy with lights on it is 19.949663391327807. The L2 distance between a cartoon of a goose and a calm lake with a few trees is 34.93465852251195. Between the two distances, the distance between a cartoon of a goose and a canopy with lights on it is smaller. Therefore, the final answer is A. a canopy with lights on it. | A. a canopy with lights on it. | multi_object_closer_to | 0011ace1c64f7919.jpg |
0011cf9a929a4e19_ac89 | Consider the real-world 3D orientations of the objects. Are a large stone building with a black roof and a white motorbike with a black leather seat facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
-0.7,
12,
38.6
],
"label": "a large stone building with a black roof"
},
{
"bbox_3d": [
0.4,
0.9,
17.8
],
"label": "a white motorbike with a black leather seat"
}
] | [
{
"front_dir": [
-0.4,
-0.3,
-0.9
],
"label": "a large stone building with a black roof",
"left_dir": [
-0.9,
0.2,
0.3
]
},
{
"front_dir": [
-1,
0.1,
-0.1
],
"label": "a white motorbike with a black leather seat",
"left_dir": [
0,
0,
1
]
}
] | B | To solve this problem, we first detect the front directions of a large stone building with a black roof and a white motorbike with a black leather seat. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a large stone building with a black roof is (-0.4, -0.3, -0.9). The front direction of a white motorbike with a black leather seat is (-1.0, 0.1, -0.1). The cosine similarity between the two front directions is 0.40, corresponding to an angle of 66.71. The angle is large, meaning that the two objects are facing very different directions. Therefore, the final answer is B. very different directions. | B. very different directions. | multi_object_same_direction | 0011cf9a929a4e19.jpg |
0011cf9a929a4e19_2788 | Consider the real-world 3D orientations of the objects. Are a red car and a motorbike with a clear windshield facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
0.9,
1,
4.8
],
"label": "a red car"
},
{
"bbox_3d": [
-4.4,
0.5,
10.4
],
"label": "a motorbike with a clear windshield"
}
] | [
{
"front_dir": [
-1,
0.1,
0.1
],
"label": "a red car",
"left_dir": [
0.1,
0,
1
]
},
{
"front_dir": [
0.5,
0,
-0.9
],
"label": "a motorbike with a clear windshield",
"left_dir": [
-0.9,
0,
-0.5
]
}
] | B | To solve this problem, we first detect the front directions of a red car and a motorbike with a clear windshield. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a red car is (-1.0, 0.1, 0.1). The front direction of a motorbike with a clear windshield is (0.5, 0.0, -0.9). The cosine similarity between the two front directions is -0.56, corresponding to an angle of 123.73. The angle is large, meaning that the two objects are facing very different directions. Therefore, the final answer is B. very different directions. | B. very different directions. | multi_object_same_direction | 0011cf9a929a4e19.jpg |
0011e949e5712f18_626f | Consider the real-world 3D locations and orientations of the objects. If I stand at a wooden bookshelf with a green box on top's position facing where it is facing, is a plastic binder with books on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
-0.2,
0.8,
0.6
],
"label": "a plastic binder with books"
},
{
"bbox_3d": [
0.3,
0.7,
0.5
],
"label": "a wooden bookshelf with a green box on top"
}
] | [
{
"front_dir": [
-0.3,
0.6,
-0.7
],
"label": "a wooden bookshelf with a green box on top",
"left_dir": [
-0.9,
0,
0.3
]
}
] | A | To solve this problem, we first determine the 3D locations of a plastic binder with books and a wooden bookshelf with a green box on top. Then we estimate the vector pointing from a wooden bookshelf with a green box on top to a plastic binder with books, as well as the left direction of a wooden bookshelf with a green box on top. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a plastic binder with books is on the left of a wooden bookshelf with a green box on top. Otherwise, a plastic binder with books is behind a wooden bookshelf with a green box on top. The 3D location of a plastic binder with books is (-0.2, 0.8, 0.6). The 3D location of a wooden bookshelf with a green box on top is (0.3, 0.7, 0.5). The vector from a wooden bookshelf with a green box on top to a plastic binder with books is hence (-0.5, 0.1, 0.1). The left direction of a wooden bookshelf with a green box on top is (-0.9, -0.0, 0.3). The cosine similarity between the vector and the left direction is 0.94, corresponding to an angle of 19.48 degrees. The angle is smaller than 90 degrees, meaning that a plastic binder with books is on the left of a wooden bookshelf with a green box on top. Therefore, the final answer is A. on the left. | A. on the left. | orientation_on_the_left | 0011e949e5712f18.jpg |
00123a36def39bf4_59e3 | Consider the real-world 3D locations and orientations of the objects. Which side of a bicycle with a black frame and a silver seat is facing a man in a white shirt and yellow shorts playing a video game? | front | left | back | right | [
{
"bbox_3d": [
1.6,
0.6,
5.4
],
"label": "a bicycle with a black frame and a silver seat"
},
{
"bbox_3d": [
-6.2,
0.8,
12.5
],
"label": "a man in a white shirt and yellow shorts playing a video game"
}
] | [
{
"front_dir": [
0.7,
-0.2,
-0.7
],
"label": "a bicycle with a black frame and a silver seat",
"left_dir": [
-0.8,
-0.3,
-0.6
]
}
] | C | To solve this problem, we first detect the 3D locations of a bicycle with a black frame and a silver seat and a man in a white shirt and yellow shorts playing a video game. Then we compute the vector pointing from a bicycle with a black frame and a silver seat to a man in a white shirt and yellow shorts playing a video game. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a bicycle with a black frame and a silver seat, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a bicycle with a black frame and a silver seat that is facing a man in a white shirt and yellow shorts playing a video game corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The 3D location of a bicycle with a black frame and a silver seat is (1.6, 0.6, 5.4). The 3D location of a man in a white shirt and yellow shorts playing a video game is (-6.2, 0.8, 12.5). The vector from a bicycle with a black frame and a silver seat to a man in a white shirt and yellow shorts playing a video game is hence (-7.9, 0.2, 7.1). The left direction of a bicycle with a black frame and a silver seat is (-0.8, -0.3, -0.6). The cosine similarity between the vector pointing to a man in a white shirt and yellow shorts playing a video game and the left direction is 0.15, corresponding to an angle of 81.33 degrees. Thus the angle between the vector pointing to a man in a white shirt and yellow shorts playing a video game and the right direction is 98.67 degrees. The front direction of a bicycle with a black frame and a silver seat is (0.7, -0.2, -0.7). The cosine similarity between the vector pointing to a man in a white shirt and yellow shorts playing a video game and the front direction is -0.97, corresponding to an angle of 166.80 degrees. Thus the angle between the vector pointing to a man in a white shirt and yellow shorts playing a video game and the back direction is 13.20 degrees. Among the four directions, the smallest angle is the back direction, with an angle of 13.20 degrees. Thus the back side of a bicycle with a black frame and a silver seat is facing the a man in a white shirt and yellow shorts playing a video game. Therefore, the final answer is C. back. | C. back. | multi_object_viewpoint_towards_object | 00123a36def39bf4.jpg |
001275e2a5ce462a_c32a | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a brown leather chair and a black chair with a wooden frame, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
2.7,
0.7,
5.9
],
"label": "a brown leather chair"
},
{
"bbox_3d": [
2.9,
0.7,
8.8
],
"label": "a black chair with a wooden frame"
}
] | [
{
"front_dir": [
-0.4,
-0.1,
-0.9
],
"label": "a brown leather chair",
"left_dir": [
-0.9,
0.1,
0.3
]
},
{
"front_dir": [
-0.3,
-0.2,
-0.9
],
"label": "a black chair with a wooden frame",
"left_dir": [
-1,
0.1,
0.2
]
}
] | A | To solve this problem, we first detect the front directions of a brown leather chair and a black chair with a wooden frame. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a brown leather chair is (-0.4, -0.1, -0.9). The front direction of a black chair with a wooden frame is (-0.3, -0.2, -0.9). The cosine similarity between the two front directions is 0.98, corresponding to an angle of 10.82. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 001275e2a5ce462a.jpg |
00129ccb99612727_e0a9 | Consider the real-world 3D locations of the objects. Are the a cat with brown fur and the a carpeted floor next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
0.2,
0.2,
0.7
],
"label": "a cat with brown fur"
},
{
"bbox_3d": [
0,
0.1,
0.5
],
"label": "a carpeted floor"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a cat with brown fur and a carpeted floor. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a cat with brown fur is (0.2, 0.2, 0.7). The 3D location of a carpeted floor is (0.0, 0.1, 0.5). The L2 distance between the two objects is 0.30. The size of the a cat with brown fur is roughly 0.59. The size of the a carpeted floor is roughly 0.80. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 00129ccb99612727.jpg |
001334773bd8d5c8_27ec | Consider the real-world 3D locations of the objects. Which object has a higher location? | a car with a black roof | a horse drawn carriage | null | null | [
{
"bbox_3d": [
1.4,
7.1,
79.5
],
"label": "a car with a black roof"
},
{
"bbox_3d": [
-3.3,
1,
11.8
],
"label": "a horse drawn carriage"
}
] | [] | A | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a car with a black roof is 11.4. The 3D height of a horse drawn carriage is 2.3. The 3D height of a car with a black roof is larger, meaning that the location of a car with a black roof is higher. Therefore, the answer is A. a horse drawn carriage. | A. a horse drawn carriage. | height_higher | 001334773bd8d5c8.jpg |
00145feb6b15d975_1b47 | Consider the real-world 3D locations of the objects. Which object has a higher location? | a white sandal | a brick sidewalk | null | null | [
{
"bbox_3d": [
0.1,
0.1,
3.2
],
"label": "a white sandal"
},
{
"bbox_3d": [
1,
0.1,
9.1
],
"label": "a brick sidewalk"
}
] | [] | B | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a white sandal is 0.3. The 3D height of a brick sidewalk is 0.6. The 3D height of a brick sidewalk is larger, meaning that the location of a brick sidewalk is higher. Therefore, the answer is B. a brick sidewalk. | B. a brick sidewalk. | height_higher | 00145feb6b15d975.jpg |
0014f79a71bd5e17_440c | Consider the real-world 3D locations and orientations of the objects. If I stand at a white plastic chair's position facing where it is facing, is a white plastic chair on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
-3.2,
1.2,
6.4
],
"label": "a white plastic chair"
},
{
"bbox_3d": [
3.4,
1.5,
6.8
],
"label": "a white plastic chair"
}
] | [
{
"front_dir": [
1,
-0.2,
0.1
],
"label": "a white plastic chair",
"left_dir": [
0.1,
0,
-1
]
},
{
"front_dir": [
-0.4,
0.2,
-0.9
],
"label": "a white plastic chair",
"left_dir": [
-0.9,
0,
0.4
]
}
] | A | To solve this problem, we first determine the 3D locations of a white plastic chair and a white plastic chair. Then we estimate the vector pointing from a white plastic chair to a white plastic chair, as well as the left direction of a white plastic chair. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a white plastic chair is on the left of a white plastic chair. Otherwise, a white plastic chair is behind a white plastic chair. The 3D location of a white plastic chair is (-3.2, 1.2, 6.4). The 3D location of a white plastic chair is (3.4, 1.5, 6.8). The vector from a white plastic chair to a white plastic chair is hence (-6.6, -0.3, -0.4). The left direction of a white plastic chair is (-0.9, -0.0, 0.4). The cosine similarity between the vector and the left direction is 0.90, corresponding to an angle of 25.20 degrees. The angle is smaller than 90 degrees, meaning that a white plastic chair is on the left of a white plastic chair. Therefore, the final answer is A. on the left. | A. on the left. | orientation_on_the_left | 0014f79a71bd5e17.jpg |
001503233549730c_021f | Consider the real-world 3D orientations of the objects. What is the relationship between the orientations of a black bicycle with a person riding it and a black bicycle with a white light, parallel of perpendicular to each other? | parallel | perpendicular | null | null | [
{
"bbox_3d": [
3.7,
-0.6,
24.7
],
"label": "a black bicycle with a person riding it"
},
{
"bbox_3d": [
2.6,
-0.7,
26.9
],
"label": "a black bicycle with a white light"
}
] | [
{
"front_dir": [
-0.1,
0,
-1
],
"label": "a black bicycle with a person riding it",
"left_dir": [
-1,
0.1,
0.1
]
},
{
"front_dir": [
0,
0,
-1
],
"label": "a black bicycle with a white light",
"left_dir": [
-1,
0.1,
0
]
}
] | A | To solve this problem, we first detect the front directions of a black bicycle with a person riding it and a black bicycle with a white light. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is closer to 0 or 180, than to 90 degrees, then the two objects are parallel to each other. Otherwise, the two objects are perpendicular to each other. The front direction of a black bicycle with a person riding it is (-0.1, 0.0, -1.0). The front direction of a black bicycle with a white light is (-0.0, 0.0, -1.0). The cosine similarity between the two front directions is 1.00, corresponding to an angle of 2.98. The angle is closer to 0 or 180, than to 90 degrees, meaning that the two objects are parallel to each other. Therefore, the final answer is A. parallel. | A. parallel. | multi_object_parallel | 001503233549730c.jpg |
001571ccc331d6b9_b053 | Consider the real-world 3D location of the objects. Which object is further away from the camera? | a stuffed animal hanging from a mirror | a grey car with red wheels | null | null | [
{
"bbox_3d": [
-0.2,
1.8,
10.4
],
"label": "a stuffed animal hanging from a mirror"
},
{
"bbox_3d": [
-0.5,
1,
4.9
],
"label": "a grey car with red wheels"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a stuffed animal hanging from a mirror and a grey car with red wheels. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a stuffed animal hanging from a mirror is (-0.2, 1.8, 10.4). The 3D location of a grey car with red wheels is (-0.5, 1.0, 4.9). The L2 distance from the camera to a stuffed animal hanging from a mirror is 10.55. The L2 distance from the camera to a grey car with red wheels is 4.98. The distance to a stuffed animal hanging from a mirror is larger. Therefore, the answer is A. a stuffed animal hanging from a mirror. | A. a stuffed animal hanging from a mirror. | location_closer_to_camera | 001571ccc331d6b9.jpg |
00166bc33fa71d0a_e667 | Consider the real-world 3D locations of the objects. Which object has a higher location? | a woman wearing a blue shirt | a red cup with a white rim | null | null | [
{
"bbox_3d": [
0.6,
1,
2.6
],
"label": "a woman wearing a blue shirt"
},
{
"bbox_3d": [
-0.3,
1,
2.2
],
"label": "a red cup with a white rim"
}
] | [] | A | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a woman wearing a blue shirt is 2.4. The 3D height of a red cup with a white rim is 1.1. The 3D height of a woman wearing a blue shirt is larger, meaning that the location of a woman wearing a blue shirt is higher. Therefore, the answer is A. a red cup with a white rim. | A. a red cup with a white rim. | height_higher | 00166bc33fa71d0a.jpg |
0016945112a29ba5_ee56 | Consider the real-world 3D locations of the objects. Is a tall tower with a green light directly above a tall building with lights on it? | yes | no | null | null | [
{
"bbox_3d": [
9.9,
41.1,
72.1
],
"label": "a tall tower with a green light"
},
{
"bbox_3d": [
5.6,
17.1,
58.1
],
"label": "a tall building with lights on it"
}
] | [] | A | To solve this problem, we first determine the 3D locations of a tall tower with a green light and a tall building with lights on it. Then we compute the vector pointing from a tall building with lights on it to a tall tower with a green light, as well as the up direction of a tall building with lights on it. We estimate the cosine similarity between the vector and the up direction, which leads to the angle between the two directions. If the angle is small, then a tall tower with a green light is directly above a tall building with lights on it. Otherwise, then a tall tower with a green light is not directly above a tall building with lights on it. The 3D location of a tall tower with a green light is (9.9, 41.1, 72.1). The 3D location of a tall building with lights on it is (5.6, 17.1, 58.1). The vector from a tall building with lights on it to a tall tower with a green light is hence (4.4, 24.0, 14.0). The up direction of a tall building with lights on it is (0.0, 1.0, 0.0). The cosine similarity between the vector and the up direction is 0.85, corresponding to an angle of 31 degrees. The angle between the vector and the up direction is small, meaning that a tall tower with a green light is directly above a tall building with lights on it. Therefore, the answer is A. yes. | A. yes. | location_above | 0016945112a29ba5.jpg |
0016c8a3e03153b6_e8b5 | Consider the real-world 3D locations and orientations of the objects. If I stand at a black wheelchair's position facing where it is facing, is a white pillar on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
0.7,
0.7,
2.9
],
"label": "a white pillar"
},
{
"bbox_3d": [
-0.5,
0.4,
1.5
],
"label": "a black wheelchair"
}
] | [
{
"front_dir": [
0.7,
0.2,
-0.7
],
"label": "a black wheelchair",
"left_dir": [
-0.7,
0.3,
-0.7
]
}
] | B | To solve this problem, we first determine the 3D locations of a white pillar and a black wheelchair. Then we estimate the vector pointing from a black wheelchair to a white pillar, as well as the left direction of a black wheelchair. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a white pillar is on the left of a black wheelchair. Otherwise, a white pillar is behind a black wheelchair. The 3D location of a white pillar is (0.7, 0.7, 2.9). The 3D location of a black wheelchair is (-0.5, 0.4, 1.5). The vector from a black wheelchair to a white pillar is hence (1.2, 0.3, 1.4). The left direction of a black wheelchair is (-0.7, 0.3, -0.7). The cosine similarity between the vector and the left direction is -0.92, corresponding to an angle of 156.63 degrees. The angle is smaller than 90 degrees, meaning that a white pillar is on the right of a black wheelchair. Therefore, the final answer is B. on the right. | B. on the right. | orientation_on_the_left | 0016c8a3e03153b6.jpg |
00170e17d0da613f_b4cf | Consider the real-world 3D location of the objects. Which object is closer to the camera? | a biscuit with a hole in the middle | a biscuit with black and white frosting | null | null | [
{
"bbox_3d": [
0.3,
0.1,
0.5
],
"label": "a biscuit with a hole in the middle"
},
{
"bbox_3d": [
0.1,
0.1,
0.4
],
"label": "a biscuit with black and white frosting"
}
] | [] | B | To solve this problem, we first estimate the 3D locations of a biscuit with a hole in the middle and a biscuit with black and white frosting. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a biscuit with a hole in the middle is (0.3, 0.1, 0.5). The 3D location of a biscuit with black and white frosting is (0.1, 0.1, 0.4). The L2 distance from the camera to a biscuit with a hole in the middle is 0.56. The L2 distance from the camera to a biscuit with black and white frosting is 0.43. The distance to a biscuit with black and white frosting is smaller. Therefore, the answer is B. a biscuit with black and white frosting. | B. a biscuit with black and white frosting. | location_closer_to_camera | 00170e17d0da613f.jpg |
0017ccd0b1865cb8_626e | Consider the real-world 3D locations and orientations of the objects. Which side of a black podium is facing a person with brown hair? | front | left | back | right | [
{
"bbox_3d": [
0.4,
1.5,
15.7
],
"label": "a black podium"
},
{
"bbox_3d": [
-0.8,
0.3,
7.5
],
"label": "a person with brown hair"
}
] | [
{
"front_dir": [
0,
-0.3,
-0.9
],
"label": "a black podium",
"left_dir": [
-1,
0.1,
-0.1
]
}
] | A | To solve this problem, we first detect the 3D locations of a black podium and a person with brown hair. Then we compute the vector pointing from a black podium to a person with brown hair. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a black podium, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a black podium that is facing a person with brown hair corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The 3D location of a black podium is (0.4, 1.5, 15.7). The 3D location of a person with brown hair is (-0.8, 0.3, 7.5). The vector from a black podium to a person with brown hair is hence (-1.2, -1.2, -8.2). The left direction of a black podium is (-1.0, 0.1, -0.1). The cosine similarity between the vector pointing to a person with brown hair and the left direction is 0.19, corresponding to an angle of 78.89 degrees. Thus the angle between the vector pointing to a person with brown hair and the right direction is 101.11 degrees. The front direction of a black podium is (0.0, -0.3, -0.9). The cosine similarity between the vector pointing to a person with brown hair and the front direction is 0.96, corresponding to an angle of 15.31 degrees. Thus the angle between the vector pointing to a person with brown hair and the back direction is 164.69 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 15.31 degrees. Thus the front side of a black podium is facing the a person with brown hair. Therefore, the final answer is A. front. | A. front. | multi_object_viewpoint_towards_object | 0017ccd0b1865cb8.jpg |
001820dafd878457_8c4a | Consider the real-world 3D location of the objects. Which object is closer to the camera? | a parking lot | a tree with green leaves | null | null | [
{
"bbox_3d": [
-2.9,
0.2,
10.7
],
"label": "a parking lot"
},
{
"bbox_3d": [
-3.7,
5.7,
25.9
],
"label": "a tree with green leaves"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a parking lot and a tree with green leaves. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a parking lot is (-2.9, 0.2, 10.7). The 3D location of a tree with green leaves is (-3.7, 5.7, 25.9). The L2 distance from the camera to a parking lot is 11.06. The L2 distance from the camera to a tree with green leaves is 26.80. The distance to a parking lot is smaller. Therefore, the answer is A. a parking lot. | A. a parking lot. | location_closer_to_camera | 001820dafd878457.jpg |
001849a427c04bdf_474e | Consider the real-world 3D locations and orientations of the objects. Which side of a black speaker with a white label is facing a conference hall with a projector screen? | front | left | back | right | [
{
"bbox_3d": [
-3.1,
2.9,
8.1
],
"label": "a black speaker with a white label"
},
{
"bbox_3d": [
-0.4,
2.4,
8.2
],
"label": "a conference hall with a projector screen"
}
] | [
{
"front_dir": [
0.4,
-0.2,
-0.9
],
"label": "a black speaker with a white label",
"left_dir": [
-0.9,
0,
-0.4
]
}
] | D | To solve this problem, we first detect the 3D locations of a black speaker with a white label and a conference hall with a projector screen. Then we compute the vector pointing from a black speaker with a white label to a conference hall with a projector screen. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a black speaker with a white label, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a black speaker with a white label that is facing a conference hall with a projector screen corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The 3D location of a black speaker with a white label is (-3.1, 2.9, 8.1). The 3D location of a conference hall with a projector screen is (-0.4, 2.4, 8.2). The vector from a black speaker with a white label to a conference hall with a projector screen is hence (2.7, -0.5, 0.1). The left direction of a black speaker with a white label is (-0.9, 0.0, -0.4). The cosine similarity between the vector pointing to a conference hall with a projector screen and the left direction is -0.90, corresponding to an angle of 154.53 degrees. Thus the angle between the vector pointing to a conference hall with a projector screen and the right direction is 25.47 degrees. The front direction of a black speaker with a white label is (0.4, -0.2, -0.9). The cosine similarity between the vector pointing to a conference hall with a projector screen and the front direction is 0.42, corresponding to an angle of 64.88 degrees. Thus the angle between the vector pointing to a conference hall with a projector screen and the back direction is 115.12 degrees. Among the four directions, the smallest angle is the right direction, with an angle of 25.47 degrees. Thus the right side of a black speaker with a white label is facing the a conference hall with a projector screen. Therefore, the final answer is D. right. | D. right. | multi_object_viewpoint_towards_object | 001849a427c04bdf.jpg |
0018645944496abb_ab91 | Consider the real-world 3D locations and orientations of the objects. Which side of a stone building with a castle-like appearance is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
-3.4,
7.7,
22.1
],
"label": "a stone building with a castle-like appearance"
}
] | [
{
"front_dir": [
-0.1,
-0.4,
-0.9
],
"label": "a stone building with a castle-like appearance",
"left_dir": [
-1,
0.1,
0.1
]
}
] | A | To solve this problem, we first estimate the 3D location of a stone building with a castle-like appearance. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a stone building with a castle-like appearance, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a stone building with a castle-like appearance that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a stone building with a castle-like appearance is (-3.4, 7.7, 22.1). The vector from a stone building with a castle-like appearance to camera is hence (3.4, -7.7, -22.1). The left direction of a stone building with a castle-like appearance is (-1.0, 0.1, 0.1). The cosine similarity between the vector pointing to camera and the left direction is -0.23, corresponding to an angle of 103.43 degrees. Thus the angle between the vector pointing to camera and the right direction is 76.57 degrees. The front direction of a stone building with a castle-like appearance is (-0.1, -0.4, -0.9). The cosine similarity between the vector pointing to camera and the front direction is 0.97, corresponding to an angle of 13.95 degrees. Thus the angle between the vector pointing to camera and the back direction is 166.05 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 13.95 degrees. Thus the front side of a stone building with a castle-like appearance is facing the camera. Therefore, the final answer is A. front. | A. front. | orientation_viewpoint | 0018645944496abb.jpg |
0018c1260fa26469_9066 | Consider the real-world 3D orientations of the objects. Are a black and white car and a car with a light on facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
4.3,
19.3,
107.2
],
"label": "a black and white car"
},
{
"bbox_3d": [
3.2,
19.3,
106.9
],
"label": "a car with a light on"
}
] | [
{
"front_dir": [
0,
-0.3,
-1
],
"label": "a black and white car",
"left_dir": [
-1,
0,
0
]
},
{
"front_dir": [
0,
-0.3,
-1
],
"label": "a car with a light on",
"left_dir": [
-1,
0,
0
]
}
] | A | To solve this problem, we first detect the front directions of a black and white car and a car with a light on. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a black and white car is (0.0, -0.3, -1.0). The front direction of a car with a light on is (0.0, -0.3, -1.0). The cosine similarity between the two front directions is 1.00, corresponding to an angle of 0.60. The angle is small, meaning that the two objects are facing same or similar directions. Therefore, the final answer is A. same or similar directions. | A. same or similar directions. | multi_object_same_direction | 0018c1260fa26469.jpg |
001905b029f4e1f1_be41 | Consider the real-world 3D location of the objects. Which object is closer to the camera? | a man in a green shirt holding up his hand | a man wearing a black and white jersey | null | null | [
{
"bbox_3d": [
-2.3,
2.8,
18.4
],
"label": "a man in a green shirt holding up his hand"
},
{
"bbox_3d": [
-1.2,
0.4,
8.8
],
"label": "a man wearing a black and white jersey"
}
] | [] | B | To solve this problem, we first estimate the 3D locations of a man in a green shirt holding up his hand and a man wearing a black and white jersey. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a man in a green shirt holding up his hand is (-2.3, 2.8, 18.4). The 3D location of a man wearing a black and white jersey is (-1.2, 0.4, 8.8). The L2 distance from the camera to a man in a green shirt holding up his hand is 18.79. The L2 distance from the camera to a man wearing a black and white jersey is 8.89. The distance to a man wearing a black and white jersey is smaller. Therefore, the answer is B. a man wearing a black and white jersey. | B. a man wearing a black and white jersey. | location_closer_to_camera | 001905b029f4e1f1.jpg |
001952f2e3bf13a5_0eb5 | Consider the real-world 3D orientations of the objects. Are a white keyboard with black keys and a white computer with a black screen facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
-0.2,
0.5,
0.9
],
"label": "a white keyboard with black keys"
},
{
"bbox_3d": [
-0.3,
0.7,
0.8
],
"label": "a white computer with a black screen"
}
] | [
{
"front_dir": [
0.3,
-0.9,
-0.3
],
"label": "a white keyboard with black keys",
"left_dir": [
-0.9,
-0.1,
-0.5
]
},
{
"front_dir": [
0.4,
-0.9,
-0.2
],
"label": "a white computer with a black screen",
"left_dir": [
-0.8,
-0.3,
-0.5
]
}
] | A | To solve this problem, we first detect the front directions of a white keyboard with black keys and a white computer with a black screen. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a white keyboard with black keys is (0.3, -0.9, -0.3). The front direction of a white computer with a black screen is (0.4, -0.9, -0.2). The cosine similarity between the two front directions is 0.98, corresponding to an angle of 10.22. The angle is small, meaning that the two objects are facing same or similar directions. Therefore, the final answer is A. same or similar directions. | A. same or similar directions. | multi_object_same_direction | 001952f2e3bf13a5.jpg |
0019b68c36d104fb_1121 | Consider the real-world 3D locations and orientations of the objects. If I stand at a piano with a brown wooden frame's position facing where it is facing, is a white car with a black hood on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
-1.3,
0.4,
3.9
],
"label": "a white car with a black hood"
},
{
"bbox_3d": [
-0.2,
0.3,
0.9
],
"label": "a piano with a brown wooden frame"
}
] | [
{
"front_dir": [
0.9,
-0.3,
0.2
],
"label": "a white car with a black hood",
"left_dir": [
0.1,
-0.2,
-1
]
},
{
"front_dir": [
-0.9,
0.3,
-0.2
],
"label": "a piano with a brown wooden frame",
"left_dir": [
-0.3,
-0.3,
0.9
]
}
] | A | To solve this problem, we first determine the 3D locations of a white car with a black hood and a piano with a brown wooden frame. Then we estimate the vector pointing from a piano with a brown wooden frame to a white car with a black hood, as well as the left direction of a piano with a brown wooden frame. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a white car with a black hood is on the left of a piano with a brown wooden frame. Otherwise, a white car with a black hood is behind a piano with a brown wooden frame. The 3D location of a white car with a black hood is (-1.3, 0.4, 3.9). The 3D location of a piano with a brown wooden frame is (-0.2, 0.3, 0.9). The vector from a piano with a brown wooden frame to a white car with a black hood is hence (-1.1, 0.1, 2.9). The left direction of a piano with a brown wooden frame is (-0.3, -0.3, 0.9). The cosine similarity between the vector and the left direction is 0.96, corresponding to an angle of 17.08 degrees. The angle is smaller than 90 degrees, meaning that a white car with a black hood is on the left of a piano with a brown wooden frame. Therefore, the final answer is A. on the left. | A. on the left. | orientation_on_the_left | 0019b68c36d104fb.jpg |
001ae662d690368e_4003 | Consider the real-world 3D locations and orientations of the objects. Which side of a white and blue trailer is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
-13.1,
-6,
95
],
"label": "a white and blue trailer"
}
] | [
{
"front_dir": [
0.2,
0.2,
-1
],
"label": "a white and blue trailer",
"left_dir": [
-1,
0.1,
-0.2
]
}
] | A | To solve this problem, we first estimate the 3D location of a white and blue trailer. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a white and blue trailer, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a white and blue trailer that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a white and blue trailer is (-13.1, -6.0, 95.0). The vector from a white and blue trailer to camera is hence (13.1, 6.0, -95.0). The left direction of a white and blue trailer is (-1.0, 0.1, -0.2). The cosine similarity between the vector pointing to camera and the left direction is 0.08, corresponding to an angle of 85.67 degrees. Thus the angle between the vector pointing to camera and the right direction is 94.33 degrees. The front direction of a white and blue trailer is (0.2, 0.2, -1.0). The cosine similarity between the vector pointing to camera and the front direction is 0.99, corresponding to an angle of 9.28 degrees. Thus the angle between the vector pointing to camera and the back direction is 170.72 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 9.28 degrees. Thus the front side of a white and blue trailer is facing the camera. Therefore, the final answer is A. front. | A. front. | orientation_viewpoint | 001ae662d690368e.jpg |
001ccf6254ebf36f_a35c | Consider the real-world 3D locations of the objects. Which is closer to a tree with green leaves, a white bench in a park or a white bench in a park? | a white bench in a park | a white bench in a park | null | null | [
{
"bbox_3d": [
2,
2.6,
5.9
],
"label": "a tree with green leaves"
},
{
"bbox_3d": [
0.6,
0.7,
4.4
],
"label": "a white bench in a park"
},
{
"bbox_3d": [
-1.4,
3.1,
6.5
],
"label": "a white bench in a park"
}
] | [] | A | To solve this problem, we first detect the 3D location of a tree with green leaves, a white bench in a park, and a white bench in a park. Then we compute the L2 distances between a tree with green leaves and a white bench in a park, and between a tree with green leaves and a white bench in a park. The object that is closer to a tree with green leaves is the one with a smaller distance. The 3D location of a tree with green leaves is (2.0, 2.6, 5.9). The 3D location of a white bench in a park is (0.6, 0.7, 4.4). The 3D location of a white bench in a park is (-1.4, 3.1, 6.5). The L2 distance between a tree with green leaves and a white bench in a park is 2.8081117999207694. The L2 distance between a tree with green leaves and a white bench in a park is 3.560825352800471. Between the two distances, the distance between a tree with green leaves and a white bench in a park is smaller. Therefore, the final answer is A. a white bench in a park. | A. a white bench in a park. | multi_object_closer_to | 001ccf6254ebf36f.jpg |
001d5341337eecf7_6dcf | Consider the real-world 3D orientations of the objects. Are a red car and a white car facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
15.4,
0.5,
33.6
],
"label": "a red car"
},
{
"bbox_3d": [
8.1,
1,
20.5
],
"label": "a white car"
}
] | [
{
"front_dir": [
-0.3,
-0.1,
-0.9
],
"label": "a red car",
"left_dir": [
-0.9,
0,
0.3
]
},
{
"front_dir": [
-0.3,
-0.1,
-1
],
"label": "a white car",
"left_dir": [
-1,
0,
0.3
]
}
] | A | To solve this problem, we first detect the front directions of a red car and a white car. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a red car is (-0.3, -0.1, -0.9). The front direction of a white car is (-0.3, -0.1, -1.0). The cosine similarity between the two front directions is 1.00, corresponding to an angle of 3.16. The angle is small, meaning that the two objects are facing same or similar directions. Therefore, the final answer is A. same or similar directions. | A. same or similar directions. | multi_object_same_direction | 001d5341337eecf7.jpg |
001d8d372f4b680d_a347 | Consider the real-world 3D locations of the objects. Is a rock with a red leaf on it directly underneath a large stone with a green netting? | yes | no | null | null | [
{
"bbox_3d": [
-1,
2,
3.8
],
"label": "a large stone with a green netting"
},
{
"bbox_3d": [
0.7,
0.6,
3.5
],
"label": "a rock with a red leaf on it"
}
] | [] | B | To solve this problem, we first determine the 3D locations of a large stone with a green netting and a rock with a red leaf on it. Then we compute the vector pointing from a rock with a red leaf on it to a large stone with a green netting, as well as the up direction of a rock with a red leaf on it. We estimate the cosine similarity between the vector and the up direction, which leads to the angle between the two directions. If the angle is small, then a large stone with a green netting is directly above a rock with a red leaf on it. Otherwise, then a large stone with a green netting is not directly above a rock with a red leaf on it. To solve the question, we first determine if a large stone with a green netting is directly above a rock with a red leaf on it. The 3D location of a large stone with a green netting is (-1.0, 2.0, 3.8). The 3D location of a rock with a red leaf on it is (0.7, 0.6, 3.5). The vector from a rock with a red leaf on it to a large stone with a green netting is hence (-1.7, 1.4, 0.2). The up direction of a rock with a red leaf on it is (0.0, 1.0, 0.0). The cosine similarity between the vector and the up direction is 0.62, corresponding to an angle of 51 degrees. The angle between the vector and the up direction is large, meaning that a large stone with a green netting is not directly above a rock with a red leaf on it. In other words, a rock with a red leaf on it is not directly underneath a large stone with a green netting. Therefore, the answer is B. no. | B. no. | location_above | 001d8d372f4b680d.jpg |
001e040e9f8a2d4f_914e | Consider the real-world 3D locations and orientations of the objects. Which side of a boat made of wood is facing the camera? | front | left | back | right | [
{
"bbox_3d": [
-0.1,
2.4,
5.3
],
"label": "a boat made of wood"
}
] | [
{
"front_dir": [
0.1,
-0.1,
-1
],
"label": "a boat made of wood",
"left_dir": [
-1,
0.1,
-0.1
]
}
] | A | To solve this problem, we first estimate the 3D location of a boat made of wood. Then we obtain the vector pointing from the object to the camera. Now we compute the angles between the vector and the left, right, front, back directions. We first compute the left direction of a boat made of wood, which leads to the angle between left direction and the vector. Immediately we also get the angle between the right direction and the vector. Similarly we compute the angles between the vector and front, back directions. The side of a boat made of wood that is facing the camera corresponds to the direction with the smallest angle. Based on the four estimated angles, we can predict the final answer. The location of a boat made of wood is (-0.1, 2.4, 5.3). The vector from a boat made of wood to camera is hence (0.1, -2.4, -5.3). The left direction of a boat made of wood is (-1.0, 0.1, -0.1). The cosine similarity between the vector pointing to camera and the left direction is 0.04, corresponding to an angle of 87.43 degrees. Thus the angle between the vector pointing to camera and the right direction is 92.57 degrees. The front direction of a boat made of wood is (0.1, -0.1, -1.0). The cosine similarity between the vector pointing to camera and the front direction is 0.96, corresponding to an angle of 16.45 degrees. Thus the angle between the vector pointing to camera and the back direction is 163.55 degrees. Among the four directions, the smallest angle is the front direction, with an angle of 16.45 degrees. Thus the front side of a boat made of wood is facing the camera. Therefore, the final answer is A. front. | A. front. | orientation_viewpoint | 001e040e9f8a2d4f.jpg |
001e5022bb22230c_a470 | Consider the real-world 3D location of the objects. Which object is closer to the camera? | a man in a black shirt | a mall with many people walking around | null | null | [
{
"bbox_3d": [
-1.2,
0.5,
3.4
],
"label": "a man in a black shirt"
},
{
"bbox_3d": [
-0.3,
3.1,
7.5
],
"label": "a mall with many people walking around"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a man in a black shirt and a mall with many people walking around. Then we estimate the L2 distances from the camera to the two objects. The object with a smaller L2 distance is the one that is closer to the camera. The 3D location of a man in a black shirt is (-1.2, 0.5, 3.4). The 3D location of a mall with many people walking around is (-0.3, 3.1, 7.5). The L2 distance from the camera to a man in a black shirt is 3.64. The L2 distance from the camera to a mall with many people walking around is 8.08. The distance to a man in a black shirt is smaller. Therefore, the answer is A. a man in a black shirt. | A. a man in a black shirt. | location_closer_to_camera | 001e5022bb22230c.jpg |
001ea8a8ca78a3bc_adce | Consider the real-world 3D locations of the objects. Which is closer to a man in a suit, a woman in a blue shirt or a man in blue shirt talking? | a woman in a blue shirt | a man in blue shirt talking | null | null | [
{
"bbox_3d": [
-0.2,
0.5,
1
],
"label": "a man in a suit"
},
{
"bbox_3d": [
-0.2,
0.5,
2
],
"label": "a woman in a blue shirt"
},
{
"bbox_3d": [
-0.3,
1.3,
3.2
],
"label": "a man in blue shirt talking"
}
] | [] | A | To solve this problem, we first detect the 3D location of a man in a suit, a woman in a blue shirt, and a man in blue shirt talking. Then we compute the L2 distances between a man in a suit and a woman in a blue shirt, and between a man in a suit and a man in blue shirt talking. The object that is closer to a man in a suit is the one with a smaller distance. The 3D location of a man in a suit is (-0.2, 0.5, 1.0). The 3D location of a woman in a blue shirt is (-0.2, 0.5, 2.0). The 3D location of a man in blue shirt talking is (-0.3, 1.3, 3.2). The L2 distance between a man in a suit and a woman in a blue shirt is 1.039652162888775. The L2 distance between a man in a suit and a man in blue shirt talking is 2.351643603667184. Between the two distances, the distance between a man in a suit and a woman in a blue shirt is smaller. Therefore, the final answer is A. a woman in a blue shirt. | A. a woman in a blue shirt. | multi_object_closer_to | 001ea8a8ca78a3bc.jpg |
001edd1f82b68837_a648 | Consider the real-world 3D locations of the objects. Is a black paddle directly above a boat with a blue and orange stripe? | yes | no | null | null | [
{
"bbox_3d": [
-1.8,
1,
9.6
],
"label": "a black paddle"
},
{
"bbox_3d": [
10.1,
-2.4,
59.8
],
"label": "a boat with a blue and orange stripe"
}
] | [] | B | To solve this problem, we first determine the 3D locations of a black paddle and a boat with a blue and orange stripe. Then we compute the vector pointing from a boat with a blue and orange stripe to a black paddle, as well as the up direction of a boat with a blue and orange stripe. We estimate the cosine similarity between the vector and the up direction, which leads to the angle between the two directions. If the angle is small, then a black paddle is directly above a boat with a blue and orange stripe. Otherwise, then a black paddle is not directly above a boat with a blue and orange stripe. The 3D location of a black paddle is (-1.8, 1.0, 9.6). The 3D location of a boat with a blue and orange stripe is (10.1, -2.4, 59.8). The vector from a boat with a blue and orange stripe to a black paddle is hence (-11.9, 3.3, -50.1). The up direction of a boat with a blue and orange stripe is (0.1, 0.9, -0.3). The cosine similarity between the vector and the up direction is 0.34, corresponding to an angle of 69 degrees. The angle between the vector and the up direction is large, meaning that a black paddle is not directly above a boat with a blue and orange stripe. Therefore, the answer is B. no. | B. no. | location_above | 001edd1f82b68837.jpg |
001ee3ac76b45faf_af06 | Consider the real-world 3D locations of the objects. Are the a tall tower with a pointed roof and the a clock with a stone wall next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
-9.8,
49.1,
90.7
],
"label": "a tall tower with a pointed roof"
},
{
"bbox_3d": [
18.6,
29.9,
69.8
],
"label": "a clock with a stone wall"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a tall tower with a pointed roof and a clock with a stone wall. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a tall tower with a pointed roof is (-9.8, 49.1, 90.7). The 3D location of a clock with a stone wall is (18.6, 29.9, 69.8). The L2 distance between the two objects is 40.12. The size of the a tall tower with a pointed roof is roughly 39.65. The size of the a clock with a stone wall is roughly 8.95. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 001ee3ac76b45faf.jpg |
001f480aa6899a21_62a2 | Consider the real-world 3D locations and orientations of the objects. If I stand at a black seat's position facing where it is facing, is a man wearing a black hat on the left or right of me? | on the left | on the right | null | null | [
{
"bbox_3d": [
0.2,
0.6,
1
],
"label": "a man wearing a black hat"
},
{
"bbox_3d": [
0.6,
0.5,
1.2
],
"label": "a black seat"
}
] | [
{
"front_dir": [
-0.4,
-0.2,
-0.9
],
"label": "a black seat",
"left_dir": [
-0.9,
0.1,
0.3
]
}
] | A | To solve this problem, we first determine the 3D locations of a man wearing a black hat and a black seat. Then we estimate the vector pointing from a black seat to a man wearing a black hat, as well as the left direction of a black seat. Next we compute the cosine similarities between the vector and the left direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a man wearing a black hat is on the left of a black seat. Otherwise, a man wearing a black hat is behind a black seat. The 3D location of a man wearing a black hat is (0.2, 0.6, 1.0). The 3D location of a black seat is (0.6, 0.5, 1.2). The vector from a black seat to a man wearing a black hat is hence (-0.4, 0.1, -0.1). The left direction of a black seat is (-0.9, 0.1, 0.3). The cosine similarity between the vector and the left direction is 0.76, corresponding to an angle of 40.76 degrees. The angle is smaller than 90 degrees, meaning that a man wearing a black hat is on the left of a black seat. Therefore, the final answer is A. on the left. | A. on the left. | orientation_on_the_left | 001f480aa6899a21.jpg |
001f4cbc9bc272e7_3884 | Consider the real-world 3D locations of the objects. Are the a bald man in a blue shirt and the a woman in a gray shirt standing next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
0.3,
1,
0.9
],
"label": "a bald man in a blue shirt"
},
{
"bbox_3d": [
-0.3,
0.9,
2.5
],
"label": "a woman in a gray shirt standing"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a bald man in a blue shirt and a woman in a gray shirt standing. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a bald man in a blue shirt is (0.3, 1.0, 0.9). The 3D location of a woman in a gray shirt standing is (-0.3, 0.9, 2.5). The L2 distance between the two objects is 1.69. The size of the a bald man in a blue shirt is roughly 0.46. The size of the a woman in a gray shirt standing is roughly 1.66. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 001f4cbc9bc272e7.jpg |
001f8d02b88a190d_c8fc | Consider the real-world 3D locations and orientations of the objects. If I stand at a black leather seat's position facing where it is facing, is a man wearing a hat in front of me or behind me? | in front of | behind | null | null | [
{
"bbox_3d": [
-0.5,
0.9,
1.1
],
"label": "a man wearing a hat"
},
{
"bbox_3d": [
0,
0.7,
1
],
"label": "a black leather seat"
}
] | [
{
"front_dir": [
-1,
0.1,
-0.1
],
"label": "a black leather seat",
"left_dir": [
-0.1,
0.1,
1
]
}
] | A | To solve this problem, we first determine the 3D locations of a man wearing a hat and a black leather seat. Then we estimate the vector pointing from a black leather seat to a man wearing a hat, as well as the front direction of a black leather seat. Next we compute the cosine similarities between the vector and the front direction, which leads to the angle between the vector and the fron direction. If the angle is smaller than 90 degrees, then a man wearing a hat is in front of a black leather seat. Otherwise, a man wearing a hat is behind a black leather seat. The 3D location of a man wearing a hat is (-0.5, 0.9, 1.1). The 3D location of a black leather seat is (-0.0, 0.7, 1.0). The vector from a black leather seat to a man wearing a hat is hence (-0.4, 0.2, 0.1). The front direction of a black leather seat is (-1.0, 0.1, -0.1). The cosine similarity between the vector and the front direction is 0.91, corresponding to an angle of 24.88 degrees. The angle is smaller than 90 degrees, meaning that a man wearing a hat is in front of a black leather seat. Therefore, the final answer is A. in front of. | A. in front of. | orientation_in_front_of | 001f8d02b88a190d.jpg |
001fb08c97bbdf6b_e9fc | Consider the real-world 3D locations of the objects. Which object has a higher location? | a man wearing a tie | a man signing a book | null | null | [
{
"bbox_3d": [
0.7,
1.1,
3.6
],
"label": "a man wearing a tie"
},
{
"bbox_3d": [
-0.3,
1,
1.4
],
"label": "a man signing a book"
}
] | [] | A | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a man wearing a tie is 2.9. The 3D height of a man signing a book is 2.0. The 3D height of a man wearing a tie is larger, meaning that the location of a man wearing a tie is higher. Therefore, the answer is A. a man signing a book. | A. a man signing a book. | height_higher | 001fb08c97bbdf6b.jpg |
00200ddc8b80344f_f56b | Consider the real-world 3D locations of the objects. Is a yellow bead with a blue design directly underneath a white bracelet? | yes | no | null | null | [
{
"bbox_3d": [
-0.3,
0.6,
1.2
],
"label": "a white bracelet"
},
{
"bbox_3d": [
0,
0.3,
0.4
],
"label": "a yellow bead with a blue design"
}
] | [] | B | To solve this problem, we first determine the 3D locations of a white bracelet and a yellow bead with a blue design. Then we compute the vector pointing from a yellow bead with a blue design to a white bracelet, as well as the up direction of a yellow bead with a blue design. We estimate the cosine similarity between the vector and the up direction, which leads to the angle between the two directions. If the angle is small, then a white bracelet is directly above a yellow bead with a blue design. Otherwise, then a white bracelet is not directly above a yellow bead with a blue design. To solve the question, we first determine if a white bracelet is directly above a yellow bead with a blue design. The 3D location of a white bracelet is (-0.3, 0.6, 1.2). The 3D location of a yellow bead with a blue design is (0.0, 0.3, 0.4). The vector from a yellow bead with a blue design to a white bracelet is hence (-0.3, 0.4, 0.8). The up direction of a yellow bead with a blue design is (0.0, 1.0, 0.0). The cosine similarity between the vector and the up direction is 0.40, corresponding to an angle of 66 degrees. The angle between the vector and the up direction is large, meaning that a white bracelet is not directly above a yellow bead with a blue design. In other words, a yellow bead with a blue design is not directly underneath a white bracelet. Therefore, the answer is B. no. | B. no. | location_above | 00200ddc8b80344f.jpg |
002114082087da38_53a6 | Consider the real-world 3D locations of the objects. Which is closer to a large stone statue, a man in brown sweater or a man in a blue shirt? | a man in brown sweater | a man in a blue shirt | null | null | [
{
"bbox_3d": [
-1,
1,
6.2
],
"label": "a large stone statue"
},
{
"bbox_3d": [
-1.8,
1,
18.8
],
"label": "a man in brown sweater"
},
{
"bbox_3d": [
-1.7,
0.7,
25.8
],
"label": "a man in a blue shirt"
}
] | [] | A | To solve this problem, we first detect the 3D location of a large stone statue, a man in brown sweater, and a man in a blue shirt. Then we compute the L2 distances between a large stone statue and a man in brown sweater, and between a large stone statue and a man in a blue shirt. The object that is closer to a large stone statue is the one with a smaller distance. The 3D location of a large stone statue is (-1.0, 1.0, 6.2). The 3D location of a man in brown sweater is (-1.8, 1.0, 18.8). The 3D location of a man in a blue shirt is (-1.7, 0.7, 25.8). The L2 distance between a large stone statue and a man in brown sweater is 12.576985680952419. The L2 distance between a large stone statue and a man in a blue shirt is 19.572082023789505. Between the two distances, the distance between a large stone statue and a man in brown sweater is smaller. Therefore, the final answer is A. a man in brown sweater. | A. a man in brown sweater. | multi_object_closer_to | 002114082087da38.jpg |
002150b620f2aa6d_96f2 | Consider the real-world 3D locations of the objects. Are the a blue car and the a statue of a man pointing upwards next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
-2.7,
0.2,
12
],
"label": "a blue car"
},
{
"bbox_3d": [
0.8,
4.1,
11.7
],
"label": "a statue of a man pointing upwards"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a blue car and a statue of a man pointing upwards. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a blue car is (-2.7, 0.2, 12.0). The 3D location of a statue of a man pointing upwards is (0.8, 4.1, 11.7). The L2 distance between the two objects is 5.19. The size of the a blue car is roughly 5.86. The size of the a statue of a man pointing upwards is roughly 2.98. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 002150b620f2aa6d.jpg |
00222ebc32fc7060_94eb | Consider the real-world 3D orientations of the objects. Are a metal stool and a wooden chair with a white table facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
-2.1,
1.2,
6.9
],
"label": "a metal stool"
},
{
"bbox_3d": [
-2.9,
1.4,
6.9
],
"label": "a wooden chair with a white table"
}
] | [
{
"front_dir": [
-0.1,
-0.1,
-1
],
"label": "a metal stool",
"left_dir": [
-1,
0.2,
0.1
]
},
{
"front_dir": [
1,
-0.3,
0
],
"label": "a wooden chair with a white table",
"left_dir": [
-0.1,
-0.4,
-0.9
]
}
] | B | To solve this problem, we first detect the front directions of a metal stool and a wooden chair with a white table. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a metal stool is (-0.1, -0.1, -1.0). The front direction of a wooden chair with a white table is (1.0, -0.3, 0.0). The cosine similarity between the two front directions is -0.12, corresponding to an angle of 96.69. The angle is large, meaning that the two objects are facing very different directions. Therefore, the final answer is B. very different directions. | B. very different directions. | multi_object_same_direction | 00222ebc32fc7060.jpg |
0022fa6e735219dc_1628 | Consider the real-world 3D locations of the objects. Which object has a lower location? | a woman wearing a black skirt | a building with a shingled roof | null | null | [
{
"bbox_3d": [
0.8,
0.3,
4
],
"label": "a woman wearing a black skirt"
},
{
"bbox_3d": [
1.1,
0.7,
3.5
],
"label": "a building with a shingled roof"
}
] | [] | A | To solve this problem, we first estimate the 3D heights of the two objects. The object with a larger height value is at a higher location. The object with a smaller height value is at a lower location. The 3D height of a woman wearing a black skirt is 0.7. The 3D height of a building with a shingled roof is 1.3. The 3D height of a building with a shingled roof is larger, meaning that the location of a building with a shingled roof is higher. In other words, the location of a woman wearing a black skirt is lower. Therefore, the answer is A. a woman wearing a black skirt. | A. a woman wearing a black skirt | height_higher | 0022fa6e735219dc.jpg |
00230cac98ad3f54_aebd | Consider the real-world 3D locations and orientations of the objects. Which object is a green motorcycle facing towards, a brick wall or the a bicycle with a round wheel? | a brick wall | a bicycle with a round wheel | null | null | [
{
"bbox_3d": [
1.1,
1,
7.5
],
"label": "a green motorcycle"
},
{
"bbox_3d": [
-1,
1.5,
3.6
],
"label": "a brick wall"
},
{
"bbox_3d": [
0.7,
0.9,
11
],
"label": "a bicycle with a round wheel"
}
] | [
{
"front_dir": [
-0.6,
-0.1,
-0.8
],
"label": "a green motorcycle",
"left_dir": [
-0.8,
0.1,
0.6
]
},
{
"front_dir": [
0,
0,
-1
],
"label": "a bicycle with a round wheel",
"left_dir": [
-1,
0,
0
]
}
] | A | To solve this problem, we first detect the 3D location of a green motorcycle, a brick wall, and a bicycle with a round wheel. Then we compute the cosine similarities between the front direction of a green motorcycle and the vectors from a green motorcycle to the other two objects. We can estimate the angles from the cosine similarities, and the object with a smaller angle is the object that a green motorcycle is facing towards. The 3D location of a green motorcycle is (1.1, 1.0, 7.5). The 3D location of a brick wall is (-1.0, 1.5, 3.6). The 3D location of a bicycle with a round wheel is (0.7, 0.9, 11.0). The front direction of a green motorcycle is (-0.6, -0.1, -0.8). First we consider if a green motorcycle is facing towards the a brick wall. The vector from a green motorcycle to a brick wall is (-2.1, 0.5, -3.8). The cosine similarity between the front direction and the vector is 0.95, corresponding to an angle of 18.12 degrees. First we consider if a green motorcycle is facing towards the a bicycle with a round wheel. The vector from a green motorcycle to a bicycle with a round wheel is (-0.3, -0.1, 3.5). The cosine similarity between the front direction and the vector is -0.69, corresponding to an angle of 133.56 degrees. We find that the angle between the front direction and a brick wall is smaller. Therefore, the final answer is A. a brick wall. | A. a brick wall. | multi_object_facing | 00230cac98ad3f54.jpg |
00240903981cd456_456b | Consider the real-world 3D locations of the objects. Are the a pink flower with purple petals and the a red roof with a white house next to each other or far away from each other? | next to each other | far away from each other | null | null | [
{
"bbox_3d": [
0.1,
0.6,
2.1
],
"label": "a pink flower with purple petals"
},
{
"bbox_3d": [
-0.2,
4.2,
9.6
],
"label": "a red roof with a white house"
}
] | [] | A | To solve this problem, we first estimate the 3D locations of a pink flower with purple petals and a red roof with a white house. Then we can compute the L2 distance between the two objects. Next we estimate the rough sizes of the two objects. If the distance between the two objects is smaller or roughly the same as the object sizes, then the two objects are next to each other. Otherwise, the two objects are far away from each other. The 3D location of a pink flower with purple petals is (0.1, 0.6, 2.1). The 3D location of a red roof with a white house is (-0.2, 4.2, 9.6). The L2 distance between the two objects is 8.29. The size of the a pink flower with purple petals is roughly 1.30. The size of the a red roof with a white house is roughly 10.07. The distance between the two objects is smaller or roughly the same as the object sizes, meaning that they are next to each other. Therefore, the answer is A. next to each other. | A. next to each other. | location_next_to | 00240903981cd456.jpg |
002435b9db3c4a03_9814 | Consider the real-world 3D orientations of the objects. Are a wooden podium with a microphone and a green bulletin board facing same or similar directions, or very different directions? | same or similar directions | very different directions | null | null | [
{
"bbox_3d": [
1.4,
1.1,
4.6
],
"label": "a wooden podium with a microphone"
},
{
"bbox_3d": [
3.1,
1.8,
5.5
],
"label": "a green bulletin board"
}
] | [
{
"front_dir": [
-0.7,
0.1,
-0.7
],
"label": "a wooden podium with a microphone",
"left_dir": [
-0.7,
-0.2,
0.7
]
},
{
"front_dir": [
-1,
0.3,
-0.2
],
"label": "a green bulletin board",
"left_dir": [
-0.2,
-0.2,
0.9
]
}
] | A | To solve this problem, we first detect the front directions of a wooden podium with a microphone and a green bulletin board. Then we compute the cosine similarities between the two front directions, and the angle between them. If the angle between the two front directions is small, then the two objects are facing same or similar directions. Otherwise, the two objects are facing very different directions. The front direction of a wooden podium with a microphone is (-0.7, 0.1, -0.7). The front direction of a green bulletin board is (-1.0, 0.3, -0.2). The cosine similarity between the two front directions is 0.84, corresponding to an angle of 33.22. The angle is small, meaning that the two objects are facing same or similar directions. Therefore, the final answer is A. same or similar directions. | A. same or similar directions. | multi_object_same_direction | 002435b9db3c4a03.jpg |
0024863edfb352bc_a57c | Consider the real-world 3D locations and orientations of the objects. Which object is a blue car facing towards, a parking lot with cars or the a blue car? | a parking lot with cars | a blue car | null | null | [
{
"bbox_3d": [
-15,
-0.6,
31.6
],
"label": "a blue car"
},
{
"bbox_3d": [
-0.3,
0.7,
9.2
],
"label": "a parking lot with cars"
},
{
"bbox_3d": [
-12.3,
-0.3,
32.6
],
"label": "a blue car"
}
] | [
{
"front_dir": [
0.5,
0,
-0.9
],
"label": "a blue car",
"left_dir": [
-0.9,
0.1,
-0.5
]
},
{
"front_dir": [
0.4,
0,
-0.9
],
"label": "a blue car",
"left_dir": [
-0.9,
0.1,
-0.4
]
}
] | A | To solve this problem, we first detect the 3D location of a blue car, a parking lot with cars, and a blue car. Then we compute the cosine similarities between the front direction of a blue car and the vectors from a blue car to the other two objects. We can estimate the angles from the cosine similarities, and the object with a smaller angle is the object that a blue car is facing towards. The 3D location of a blue car is (-15.0, -0.6, 31.6). The 3D location of a parking lot with cars is (-0.3, 0.7, 9.2). The 3D location of a blue car is (-12.3, -0.3, 32.6). The front direction of a blue car is (0.5, 0.0, -0.9). First we consider if a blue car is facing towards the a parking lot with cars. The vector from a blue car to a parking lot with cars is (14.8, 1.3, -22.4). The cosine similarity between the front direction and the vector is 1.00, corresponding to an angle of 4.45 degrees. First we consider if a blue car is facing towards the a blue car. The vector from a blue car to a blue car is (2.7, 0.3, 1.0). The cosine similarity between the front direction and the vector is 0.16, corresponding to an angle of 80.65 degrees. We find that the angle between the front direction and a parking lot with cars is smaller. Therefore, the final answer is A. a parking lot with cars. | A. a parking lot with cars. | multi_object_facing | 0024863edfb352bc.jpg |
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 72
Size of downloaded dataset files:
24 MB
Size of the auto-converted Parquet files:
24 MB
Number of rows:
60,000
Models trained or fine-tuned on ccvl/OpenImages_3DSR_feb27_60k