Spaces:
Sleeping
Sleeping
Upload folder using huggingface_hub
Browse files- app.py +3 -1
- inference.py +31 -54
- u2net/inference.py +11 -7
app.py
CHANGED
@@ -30,7 +30,8 @@ def load_model_without_module(model, model_path):
|
|
30 |
name = k[7:] if k.startswith('module.') else k
|
31 |
new_state_dict[name] = v
|
32 |
model.load_state_dict(new_state_dict)
|
33 |
-
|
|
|
34 |
model = VGG_19().to(device).eval()
|
35 |
for param in model.parameters():
|
36 |
param.requires_grad = False
|
@@ -43,6 +44,7 @@ style_options = {' '.join(style_file.split('.')[0].split('_')): f'./style_images
|
|
43 |
lrs = np.logspace(np.log10(0.001), np.log10(0.1), 10).tolist()
|
44 |
img_size = 512
|
45 |
|
|
|
46 |
cached_style_features = {}
|
47 |
for style_name, style_img_path in style_options.items():
|
48 |
style_img = preprocess_img_from_path(style_img_path, img_size)[0].to(device)
|
|
|
30 |
name = k[7:] if k.startswith('module.') else k
|
31 |
new_state_dict[name] = v
|
32 |
model.load_state_dict(new_state_dict)
|
33 |
+
|
34 |
+
# load models
|
35 |
model = VGG_19().to(device).eval()
|
36 |
for param in model.parameters():
|
37 |
param.requires_grad = False
|
|
|
44 |
lrs = np.logspace(np.log10(0.001), np.log10(0.1), 10).tolist()
|
45 |
img_size = 512
|
46 |
|
47 |
+
# store style(s) features
|
48 |
cached_style_features = {}
|
49 |
for style_name, style_img_path in style_options.items():
|
50 |
style_img = preprocess_img_from_path(style_img_path, img_size)[0].to(device)
|
inference.py
CHANGED
@@ -1,37 +1,24 @@
|
|
1 |
-
import os
|
2 |
-
from tqdm import tqdm
|
3 |
-
|
4 |
import torch
|
5 |
import torch.optim as optim
|
6 |
import torch.nn.functional as F
|
7 |
from torchvision.transforms.functional import gaussian_blur
|
|
|
8 |
|
9 |
-
def
|
10 |
-
|
11 |
-
|
12 |
-
return
|
13 |
|
14 |
-
def
|
15 |
-
content_loss =
|
16 |
-
style_loss =
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
masked_gf = gf * blurred_bg_mask
|
25 |
-
masked_sf = sf * blurred_bg_mask
|
26 |
-
G = _gram_matrix(masked_gf)
|
27 |
-
A = _gram_matrix(masked_sf)
|
28 |
-
else:
|
29 |
-
G = _gram_matrix(gf)
|
30 |
-
A = _gram_matrix(sf)
|
31 |
-
style_loss += w_l * F.mse_loss(G, A)
|
32 |
-
|
33 |
-
total_loss = alpha * content_loss + beta * style_loss
|
34 |
-
return content_loss, style_loss, total_loss
|
35 |
|
36 |
def inference(
|
37 |
*,
|
@@ -41,7 +28,7 @@ def inference(
|
|
41 |
content_image_norm,
|
42 |
style_features,
|
43 |
apply_to_background,
|
44 |
-
lr,
|
45 |
iterations=101,
|
46 |
optim_caller=optim.AdamW,
|
47 |
alpha=1,
|
@@ -49,43 +36,33 @@ def inference(
|
|
49 |
):
|
50 |
generated_image = content_image.clone().requires_grad_(True)
|
51 |
optimizer = optim_caller([generated_image], lr=lr)
|
52 |
-
min_losses = [float('inf')] * iterations
|
53 |
|
54 |
with torch.no_grad():
|
55 |
content_features = model(content_image)
|
56 |
-
|
57 |
-
|
58 |
if apply_to_background:
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
for cf in content_features:
|
66 |
-
_, _, h_i, w_i = cf.shape
|
67 |
-
bg_mask = F.interpolate(background_mask.unsqueeze(1), size=(h_i, w_i), mode='bilinear', align_corners=False)
|
68 |
-
resized_bg_masks.append(bg_mask)
|
69 |
|
70 |
-
def closure(
|
71 |
optimizer.zero_grad()
|
72 |
generated_features = model(generated_image)
|
73 |
-
content_loss, style_loss, total_loss =
|
74 |
-
generated_features, content_features, style_features,
|
75 |
)
|
76 |
total_loss.backward()
|
77 |
-
|
78 |
-
# log loss
|
79 |
-
min_losses[iter] = min(min_losses[iter], total_loss.item())
|
80 |
-
|
81 |
return total_loss
|
82 |
|
83 |
-
for
|
84 |
-
optimizer.step(
|
85 |
-
|
86 |
if apply_to_background:
|
87 |
with torch.no_grad():
|
88 |
-
|
89 |
-
generated_image.data
|
90 |
|
91 |
return generated_image
|
|
|
|
|
|
|
|
|
1 |
import torch
|
2 |
import torch.optim as optim
|
3 |
import torch.nn.functional as F
|
4 |
from torchvision.transforms.functional import gaussian_blur
|
5 |
+
from tqdm import tqdm
|
6 |
|
7 |
+
def gram_matrix(feature):
|
8 |
+
b, c, h, w = feature.size()
|
9 |
+
feature = feature.view(b * c, h * w)
|
10 |
+
return feature @ feature.t()
|
11 |
|
12 |
+
def compute_loss(generated, content, style, bg_masks, alpha, beta):
|
13 |
+
content_loss = sum(F.mse_loss(gf, cf) for gf, cf in zip(generated, content))
|
14 |
+
style_loss = sum(
|
15 |
+
F.mse_loss(
|
16 |
+
gram_matrix(gf * bg) if bg is not None else gram_matrix(gf),
|
17 |
+
gram_matrix(sf * bg) if bg is not None else gram_matrix(sf),
|
18 |
+
) / len(generated)
|
19 |
+
for gf, sf, bg in zip(generated, style, bg_masks or [None] * len(generated))
|
20 |
+
)
|
21 |
+
return alpha * content_loss, beta * style_loss, alpha * content_loss + beta * style_loss
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
def inference(
|
24 |
*,
|
|
|
28 |
content_image_norm,
|
29 |
style_features,
|
30 |
apply_to_background,
|
31 |
+
lr=5e-2,
|
32 |
iterations=101,
|
33 |
optim_caller=optim.AdamW,
|
34 |
alpha=1,
|
|
|
36 |
):
|
37 |
generated_image = content_image.clone().requires_grad_(True)
|
38 |
optimizer = optim_caller([generated_image], lr=lr)
|
|
|
39 |
|
40 |
with torch.no_grad():
|
41 |
content_features = model(content_image)
|
42 |
+
bg_masks = None
|
43 |
+
|
44 |
if apply_to_background:
|
45 |
+
seg_output = torch.sigmoid(sod_model(content_image_norm)[0])
|
46 |
+
bg_mask = (seg_output <= 0.7).float()
|
47 |
+
bg_masks = [
|
48 |
+
F.interpolate(bg_mask.unsqueeze(1), size=cf.shape[2:], mode='bilinear', align_corners=False)
|
49 |
+
for cf in content_features
|
50 |
+
]
|
|
|
|
|
|
|
|
|
51 |
|
52 |
+
def closure():
|
53 |
optimizer.zero_grad()
|
54 |
generated_features = model(generated_image)
|
55 |
+
content_loss, style_loss, total_loss = compute_loss(
|
56 |
+
generated_features, content_features, style_features, bg_masks, alpha, beta
|
57 |
)
|
58 |
total_loss.backward()
|
|
|
|
|
|
|
|
|
59 |
return total_loss
|
60 |
|
61 |
+
for _ in tqdm(range(iterations)):
|
62 |
+
optimizer.step(closure)
|
|
|
63 |
if apply_to_background:
|
64 |
with torch.no_grad():
|
65 |
+
fg_mask = F.interpolate(1 - bg_masks[0], size=generated_image.shape[2:], mode='nearest')
|
66 |
+
generated_image.data.mul_(1 - fg_mask).add_(content_image.data * fg_mask)
|
67 |
|
68 |
return generated_image
|
u2net/inference.py
CHANGED
@@ -9,19 +9,22 @@ from matplotlib.gridspec import GridSpec
|
|
9 |
|
10 |
from model import U2Net
|
11 |
|
12 |
-
|
|
|
|
|
|
|
13 |
|
14 |
def preprocess_image(image_path):
|
15 |
img = Image.open(image_path).convert('RGB')
|
16 |
preprocess = transforms.Compose([
|
17 |
-
transforms.Resize((512, 512)),
|
18 |
transforms.ToTensor(),
|
19 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
20 |
])
|
21 |
img = preprocess(img).unsqueeze(0).to(device)
|
22 |
return img
|
23 |
|
24 |
-
def run_inference(model, image_path, threshold=
|
25 |
input_img = preprocess_image(image_path)
|
26 |
with torch.no_grad():
|
27 |
d1, *_ = model(input_img)
|
@@ -47,15 +50,16 @@ def overlay_segmentation(original_image, binary_mask, alpha=0.5):
|
|
47 |
|
48 |
if __name__ == '__main__':
|
49 |
# ---
|
50 |
-
model_path = '
|
51 |
-
|
|
|
52 |
# ---
|
53 |
model = U2Net().to(device)
|
54 |
model = nn.DataParallel(model)
|
55 |
model.load_state_dict(load_file(model_path, device=device.type))
|
56 |
model.eval()
|
57 |
|
58 |
-
mask = run_inference(model, image_path
|
59 |
mask_with_threshold = run_inference(model, image_path, threshold=0.7)
|
60 |
|
61 |
fig = plt.figure(figsize=(10, 10))
|
@@ -74,4 +78,4 @@ if __name__ == '__main__':
|
|
74 |
ax.axis('off')
|
75 |
|
76 |
plt.subplots_adjust(left=0, right=1, top=1, bottom=0)
|
77 |
-
plt.savefig('inference-output.jpg', format='jpg', bbox_inches='tight', pad_inches=0)
|
|
|
9 |
|
10 |
from model import U2Net
|
11 |
|
12 |
+
if torch.cuda.is_available(): device = 'cuda'
|
13 |
+
elif torch.backends.mps.is_available(): device = 'mps'
|
14 |
+
else: device = 'cpu'
|
15 |
+
device = torch.device(device)
|
16 |
|
17 |
def preprocess_image(image_path):
|
18 |
img = Image.open(image_path).convert('RGB')
|
19 |
preprocess = transforms.Compose([
|
20 |
+
transforms.Resize((512, 512), interpolation=transforms.InterpolationMode.BILINEAR),
|
21 |
transforms.ToTensor(),
|
22 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
23 |
])
|
24 |
img = preprocess(img).unsqueeze(0).to(device)
|
25 |
return img
|
26 |
|
27 |
+
def run_inference(model, image_path, threshold=None):
|
28 |
input_img = preprocess_image(image_path)
|
29 |
with torch.no_grad():
|
30 |
d1, *_ = model(input_img)
|
|
|
50 |
|
51 |
if __name__ == '__main__':
|
52 |
# ---
|
53 |
+
model_path = '../testing/u2net-duts-msra.safetensors'
|
54 |
+
filename = input('Filename: ')
|
55 |
+
image_path = f'../content_images/{filename}'
|
56 |
# ---
|
57 |
model = U2Net().to(device)
|
58 |
model = nn.DataParallel(model)
|
59 |
model.load_state_dict(load_file(model_path, device=device.type))
|
60 |
model.eval()
|
61 |
|
62 |
+
mask = run_inference(model, image_path)
|
63 |
mask_with_threshold = run_inference(model, image_path, threshold=0.7)
|
64 |
|
65 |
fig = plt.figure(figsize=(10, 10))
|
|
|
78 |
ax.axis('off')
|
79 |
|
80 |
plt.subplots_adjust(left=0, right=1, top=1, bottom=0)
|
81 |
+
plt.savefig('../testing/inference-output.jpg', format='jpg', bbox_inches='tight', pad_inches=0)
|