Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -2,31 +2,43 @@ import gradio as gr
|
|
2 |
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
|
3 |
from PIL import Image
|
4 |
import torch
|
|
|
5 |
import matplotlib.pyplot as plt
|
6 |
-
import spaces
|
7 |
|
8 |
# Load TrOCR model
|
9 |
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-large-handwritten")
|
10 |
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-large-handwritten")
|
11 |
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
def recognize_text(image):
|
14 |
try:
|
15 |
-
# Convert image to RGB if it's not already
|
16 |
-
image = image.convert("RGB")
|
17 |
-
print("Image converted to RGB.")
|
18 |
-
|
19 |
# Preprocess the image
|
20 |
-
pixel_values =
|
21 |
print("Image preprocessed. Pixel values shape:", pixel_values.shape)
|
22 |
|
23 |
# Visualize preprocessed image
|
24 |
-
|
25 |
-
plt.title("Preprocessed Image")
|
26 |
-
plt.show()
|
27 |
|
28 |
# Generate text from the image
|
29 |
-
with torch.no_grad():
|
30 |
generated_ids = model.generate(pixel_values)
|
31 |
print("Generated IDs:", generated_ids)
|
32 |
|
|
|
2 |
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
|
3 |
from PIL import Image
|
4 |
import torch
|
5 |
+
from torchvision import transforms
|
6 |
import matplotlib.pyplot as plt
|
|
|
7 |
|
8 |
# Load TrOCR model
|
9 |
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-large-handwritten")
|
10 |
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-large-handwritten")
|
11 |
|
12 |
+
def preprocess_image(image):
|
13 |
+
# Convert image to RGB
|
14 |
+
image = image.convert("RGB")
|
15 |
+
|
16 |
+
# Resize and normalize the image to [0, 1]
|
17 |
+
transform = transforms.Compose([
|
18 |
+
transforms.Resize((384, 384)), # Resize to the expected input size
|
19 |
+
transforms.ToTensor(), # Convert to tensor and scale to [0, 1]
|
20 |
+
])
|
21 |
+
pixel_values = transform(image).unsqueeze(0) # Add batch dimension
|
22 |
+
return pixel_values
|
23 |
+
|
24 |
+
def visualize_image(pixel_values):
|
25 |
+
# Convert tensor to numpy array and permute dimensions for visualization
|
26 |
+
image = pixel_values.squeeze().permute(1, 2, 0).numpy()
|
27 |
+
plt.imshow(image)
|
28 |
+
plt.title("Preprocessed Image")
|
29 |
+
plt.show()
|
30 |
+
|
31 |
def recognize_text(image):
|
32 |
try:
|
|
|
|
|
|
|
|
|
33 |
# Preprocess the image
|
34 |
+
pixel_values = preprocess_image(image)
|
35 |
print("Image preprocessed. Pixel values shape:", pixel_values.shape)
|
36 |
|
37 |
# Visualize preprocessed image
|
38 |
+
visualize_image(pixel_values)
|
|
|
|
|
39 |
|
40 |
# Generate text from the image
|
41 |
+
with torch.no_grad():
|
42 |
generated_ids = model.generate(pixel_values)
|
43 |
print("Generated IDs:", generated_ids)
|
44 |
|