update README.md
Browse files
README.md
CHANGED
@@ -65,24 +65,67 @@ This dataset contains aesthetic annotations for images. The annotations cover 18
|
|
65 |
|
66 |
Each image in the dataset is annotated with the following attributes:
|
67 |
|
68 |
-
1
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
|
87 |
### Example: Scene Richness (richness)
|
88 |
- **2:** Very rich
|
@@ -105,11 +148,13 @@ The `annotation` feature contains scores across 18 different dimensions of image
|
|
105 |
### Meta Result
|
106 |
The `meta_result` feature transforms multi-choice questions into a series of binary judgments. For example, for the `richness` dimension:
|
107 |
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
|
|
|
|
113 |
|
114 |
Each element in the binary array represents a yes/no answer to a specific aspect of the assessment. For detailed questions corresponding to these binary judgments, please refer to the `meta_qa_en.txt` file.
|
115 |
|
@@ -123,4 +168,5 @@ The `meta_mask` feature is used for balanced sampling during model training:
|
|
123 |
We provide `extract.py` for processing the dataset into JSONL format. The script can optionally extract the balanced positive/negative QA pairs used in VisionReward training by processing `meta_result` and `meta_mask` fields.
|
124 |
|
125 |
```bash
|
126 |
-
python extract.py [--save_imgs] [--process_qa]
|
|
|
|
65 |
|
66 |
Each image in the dataset is annotated with the following attributes:
|
67 |
|
68 |
+
<table border="1" style="border-collapse: collapse;">
|
69 |
+
<tr>
|
70 |
+
<th>Dimension</th>
|
71 |
+
<th>Attribute</th>
|
72 |
+
</tr>
|
73 |
+
<tr>
|
74 |
+
<td rowspan="5">Composition</td>
|
75 |
+
<td>Symmetry</td>
|
76 |
+
</tr>
|
77 |
+
<tr>
|
78 |
+
<td>Object pairing</td>
|
79 |
+
</tr>
|
80 |
+
<tr>
|
81 |
+
<td>Main object</td>
|
82 |
+
</tr>
|
83 |
+
<tr>
|
84 |
+
<td>Richness</td>
|
85 |
+
</tr>
|
86 |
+
<tr>
|
87 |
+
<td>Background</td>
|
88 |
+
</tr>
|
89 |
+
<tr>
|
90 |
+
<td rowspan="5">Quality</td>
|
91 |
+
<td>Clarity</td>
|
92 |
+
</tr>
|
93 |
+
<tr>
|
94 |
+
<td>Color Brightness</td>
|
95 |
+
</tr>
|
96 |
+
<tr>
|
97 |
+
<td>Color Aesthetic</td>
|
98 |
+
</tr>
|
99 |
+
<tr>
|
100 |
+
<td>Lighting Distinction</td>
|
101 |
+
</tr>
|
102 |
+
<tr>
|
103 |
+
<td>Lighting Aesthetic</td>
|
104 |
+
</tr>
|
105 |
+
<tr>
|
106 |
+
<td rowspan="5">Fidelity</td>
|
107 |
+
<td>Detail realism</td>
|
108 |
+
</tr>
|
109 |
+
<tr>
|
110 |
+
<td>Detail refinement</td>
|
111 |
+
</tr>
|
112 |
+
<tr>
|
113 |
+
<td>Body</td>
|
114 |
+
</tr>
|
115 |
+
<tr>
|
116 |
+
<td>Face</td>
|
117 |
+
</tr>
|
118 |
+
<tr>
|
119 |
+
<td>Hands</td>
|
120 |
+
</tr>
|
121 |
+
<tr>
|
122 |
+
<td rowspan="2">Safety & Emotion</td>
|
123 |
+
<td>Emotion</td>
|
124 |
+
</tr>
|
125 |
+
<tr>
|
126 |
+
<td>Safety</td>
|
127 |
+
</tr>
|
128 |
+
</table>
|
129 |
|
130 |
### Example: Scene Richness (richness)
|
131 |
- **2:** Very rich
|
|
|
148 |
### Meta Result
|
149 |
The `meta_result` feature transforms multi-choice questions into a series of binary judgments. For example, for the `richness` dimension:
|
150 |
|
151 |
+
| Score | Is the image very rich? | Is the image rich? | Is the image not monotonous? | Is the image not empty? |
|
152 |
+
|-------|------------------------|-------------------|---------------------------|----------------------|
|
153 |
+
| 2 | 1 | 1 | 1 | 1 |
|
154 |
+
| 1 | 0 | 1 | 1 | 1 |
|
155 |
+
| 0 | 0 | 0 | 1 | 1 |
|
156 |
+
| -1 | 0 | 0 | 0 | 1 |
|
157 |
+
| -2 | 0 | 0 | 0 | 0 |
|
158 |
|
159 |
Each element in the binary array represents a yes/no answer to a specific aspect of the assessment. For detailed questions corresponding to these binary judgments, please refer to the `meta_qa_en.txt` file.
|
160 |
|
|
|
168 |
We provide `extract.py` for processing the dataset into JSONL format. The script can optionally extract the balanced positive/negative QA pairs used in VisionReward training by processing `meta_result` and `meta_mask` fields.
|
169 |
|
170 |
```bash
|
171 |
+
python extract.py [--save_imgs] [--process_qa]
|
172 |
+
```
|