Spaces:
Running
on
Zero
Running
on
Zero
Change to tab_batch to take dynamic number of images as the input.
Browse files
app.py
CHANGED
@@ -36,8 +36,24 @@ model.eval()
|
|
36 |
|
37 |
|
38 |
@spaces.GPU
|
39 |
-
def pred_maps(
|
40 |
-
images
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
image_shapes = [image.shape[:2] for image in images]
|
42 |
images = [Image.fromarray(image) for image in images]
|
43 |
|
@@ -50,30 +66,33 @@ def pred_maps(image_1, image_2, image_3, image_4):
|
|
50 |
with torch.no_grad():
|
51 |
scaled_preds_tensor = model(images_proc.to(device))[-1]
|
52 |
preds = []
|
53 |
-
for image_shape, pred_tensor in zip(image_shapes, scaled_preds_tensor):
|
54 |
if device == 'cuda':
|
55 |
pred_tensor = pred_tensor.cpu()
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
|
|
63 |
|
64 |
|
65 |
-
N = 4
|
66 |
# examples = [[_] for _ in glob('example_images/butterfly/*')][:N]
|
67 |
|
68 |
-
|
69 |
-
opt = [gr.Image(width=600, height=300) for _ in range(N)]
|
70 |
-
demo = gr.Interface(
|
71 |
fn=pred_maps,
|
72 |
-
inputs=
|
73 |
-
outputs=
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
|
|
|
|
|
|
|
|
78 |
)
|
79 |
demo.launch(debug=True)
|
|
|
36 |
|
37 |
|
38 |
@spaces.GPU
|
39 |
+
def pred_maps(images):
|
40 |
+
assert (images is not None), 'AssertionError: images cannot be None.'
|
41 |
+
# For tab_batch
|
42 |
+
save_paths = []
|
43 |
+
save_dir = 'preds-GCoNet_plus'
|
44 |
+
if not os.path.exists(save_dir):
|
45 |
+
os.makedirs(save_dir)
|
46 |
+
|
47 |
+
image_array_lst = []
|
48 |
+
for idx_image, image_src in enumerate(images):
|
49 |
+
save_paths.append(os.path.join(save_dir, "{}.png".format(os.path.splitext(os.path.basename(image_src))[0])))
|
50 |
+
if isinstance(image_src, str):
|
51 |
+
image = np.array(Image.open(image_src))
|
52 |
+
else:
|
53 |
+
image = image_src
|
54 |
+
image_array_lst.append(image)
|
55 |
+
images = image_array_lst
|
56 |
+
|
57 |
image_shapes = [image.shape[:2] for image in images]
|
58 |
images = [Image.fromarray(image) for image in images]
|
59 |
|
|
|
66 |
with torch.no_grad():
|
67 |
scaled_preds_tensor = model(images_proc.to(device))[-1]
|
68 |
preds = []
|
69 |
+
for image_shape, pred_tensor, save_path in zip(image_shapes, scaled_preds_tensor, save_paths):
|
70 |
if device == 'cuda':
|
71 |
pred_tensor = pred_tensor.cpu()
|
72 |
+
pred_tensor = torch.nn.functional.interpolate(pred_tensor.unsqueeze(0), size=image_shape, mode='bilinear', align_corners=True).squeeze().numpy()
|
73 |
+
cv2.imwrite(save_path, pred_tensor)
|
74 |
+
|
75 |
+
zip_file_path = os.path.join(save_dir, "{}.zip".format(save_dir))
|
76 |
+
with zipfile.ZipFile(zip_file_path, 'w') as zipf:
|
77 |
+
for file in save_paths:
|
78 |
+
zipf.write(file, os.path.basename(file))
|
79 |
+
return save_paths, zip_file_path
|
80 |
|
81 |
|
82 |
+
# N = 4
|
83 |
# examples = [[_] for _ in glob('example_images/butterfly/*')][:N]
|
84 |
|
85 |
+
tab_batch = gr.Interface(
|
|
|
|
|
86 |
fn=pred_maps,
|
87 |
+
inputs=gr.File(label="Upload multiple images in a group", type="filepath", file_count="multiple"),
|
88 |
+
outputs=[gr.Gallery(label="GCoNet+'s predictions"), gr.File(label="Download predicted maps.")],
|
89 |
+
api_name="batch",
|
90 |
+
description='Upload pictures, most of which contain salient objects of the same class. Our demo will give you the binary maps of these co-salient objects :)',
|
91 |
+
)
|
92 |
+
|
93 |
+
demo = gr.TabbedInterface(
|
94 |
+
[tab_batch],
|
95 |
+
['batch'],
|
96 |
+
title="Online demo for `GCoNet+: A Stronger Group Collaborative Co-Salient Object Detector (T-PAMI 2023)`",
|
97 |
)
|
98 |
demo.launch(debug=True)
|