Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -13,53 +13,63 @@ model = SamModel.from_pretrained("facebook/sam-vit-base").to(device)
|
|
13 |
processor = SamProcessor.from_pretrained("facebook/sam-vit-base")
|
14 |
|
15 |
def segment_image(input_image, segment_anything):
|
16 |
-
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
# Store original size
|
23 |
-
original_size = input_image.size
|
24 |
-
|
25 |
-
if segment_anything:
|
26 |
-
# Segment everything in the image
|
27 |
-
inputs = processor(input_image, return_tensors="pt").to(device)
|
28 |
-
else:
|
29 |
-
# Use the center of the image as a point prompt
|
30 |
-
width, height = input_image.size
|
31 |
-
center_point = [[width // 2, height // 2]]
|
32 |
-
inputs = processor(input_image, input_points=[center_point], return_tensors="pt").to(device)
|
33 |
-
|
34 |
-
# Generate masks
|
35 |
-
with torch.no_grad():
|
36 |
-
outputs = model(**inputs)
|
37 |
-
|
38 |
-
# Post-process masks
|
39 |
-
masks = processor.image_processor.post_process_masks(
|
40 |
-
outputs.pred_masks.cpu(),
|
41 |
-
inputs["original_sizes"].cpu(),
|
42 |
-
inputs["reshaped_input_sizes"].cpu()
|
43 |
-
)
|
44 |
-
|
45 |
-
# Convert mask to numpy array and resize to match original image
|
46 |
-
if segment_anything:
|
47 |
-
# Combine all masks
|
48 |
-
combined_mask = np.any(masks[0].numpy() > 0.5, axis=0)
|
49 |
-
else:
|
50 |
-
# Use the first mask
|
51 |
-
combined_mask = masks[0][0].numpy() > 0.5
|
52 |
-
|
53 |
-
# Resize mask to match original image size
|
54 |
-
combined_mask = cv2.resize(combined_mask.astype(np.uint8), original_size[::-1]) > 0
|
55 |
-
|
56 |
-
# Overlay the mask on the original image
|
57 |
-
result_image = np.array(input_image)
|
58 |
-
mask_rgb = np.zeros_like(result_image)
|
59 |
-
mask_rgb[combined_mask] = [255, 0, 0] # Red color for the mask
|
60 |
-
result_image = cv2.addWeighted(result_image, 1, mask_rgb, 0.5, 0)
|
61 |
-
|
62 |
-
return result_image, "Segmentation completed successfully."
|
63 |
|
64 |
# Create Gradio interface
|
65 |
iface = gr.Interface(
|
|
|
13 |
processor = SamProcessor.from_pretrained("facebook/sam-vit-base")
|
14 |
|
15 |
def segment_image(input_image, segment_anything):
|
16 |
+
try:
|
17 |
+
if input_image is None:
|
18 |
+
return None, "Please upload an image before submitting."
|
19 |
+
|
20 |
+
# Convert input_image to PIL Image
|
21 |
+
input_image = Image.fromarray(input_image).convert("RGB")
|
22 |
+
|
23 |
+
# Store original size
|
24 |
+
original_size = input_image.size
|
25 |
+
if not original_size or 0 in original_size:
|
26 |
+
return None, "Invalid image size. Please upload a different image."
|
27 |
+
|
28 |
+
if segment_anything:
|
29 |
+
# Segment everything in the image
|
30 |
+
inputs = processor(input_image, return_tensors="pt").to(device)
|
31 |
+
else:
|
32 |
+
# Use the center of the image as a point prompt
|
33 |
+
width, height = original_size
|
34 |
+
center_point = [[width // 2, height // 2]]
|
35 |
+
inputs = processor(input_image, input_points=[center_point], return_tensors="pt").to(device)
|
36 |
+
|
37 |
+
# Generate masks
|
38 |
+
with torch.no_grad():
|
39 |
+
outputs = model(**inputs)
|
40 |
+
|
41 |
+
# Post-process masks
|
42 |
+
masks = processor.image_processor.post_process_masks(
|
43 |
+
outputs.pred_masks.cpu(),
|
44 |
+
inputs["original_sizes"].cpu(),
|
45 |
+
inputs["reshaped_input_sizes"].cpu()
|
46 |
+
)
|
47 |
+
|
48 |
+
# Convert mask to numpy array and resize to match original image
|
49 |
+
if segment_anything:
|
50 |
+
# Combine all masks
|
51 |
+
combined_mask = np.any(masks[0].numpy() > 0.5, axis=0)
|
52 |
+
else:
|
53 |
+
# Use the first mask
|
54 |
+
combined_mask = masks[0][0].numpy() > 0.5
|
55 |
+
|
56 |
+
# Ensure mask is 2D
|
57 |
+
if combined_mask.ndim > 2:
|
58 |
+
combined_mask = combined_mask.squeeze()
|
59 |
+
|
60 |
+
# Resize mask to match original image size
|
61 |
+
combined_mask = cv2.resize(combined_mask.astype(np.uint8), (original_size[0], original_size[1])) > 0
|
62 |
+
|
63 |
+
# Overlay the mask on the original image
|
64 |
+
result_image = np.array(input_image)
|
65 |
+
mask_rgb = np.zeros_like(result_image)
|
66 |
+
mask_rgb[combined_mask] = [255, 0, 0] # Red color for the mask
|
67 |
+
result_image = cv2.addWeighted(result_image, 1, mask_rgb, 0.5, 0)
|
68 |
+
|
69 |
+
return result_image, "Segmentation completed successfully."
|
70 |
|
71 |
+
except Exception as e:
|
72 |
+
return None, f"An error occurred: {str(e)}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
# Create Gradio interface
|
75 |
iface = gr.Interface(
|