Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -6,130 +6,10 @@ from huggingface_hub import hf_hub_download
|
|
6 |
import torch
|
7 |
import json
|
8 |
from omegaconf import OmegaConf
|
9 |
-
from model import Generator
|
10 |
-
|
11 |
-
|
12 |
-
class BaseGenerator(ABC, nn.Module):
|
13 |
-
def __init__(self, channels: int = 3):
|
14 |
-
super().__init__()
|
15 |
-
self.channels = channels
|
16 |
-
|
17 |
-
@abstractmethod
|
18 |
-
def forward(self, x: Tensor) -> Tensor:
|
19 |
-
pass
|
20 |
-
|
21 |
-
|
22 |
-
class Generator(BaseGenerator):
|
23 |
-
def __init__(self, cfg: DictConfig):
|
24 |
-
super().__init__(cfg.channels)
|
25 |
-
self.cfg = cfg
|
26 |
-
self.model = self._construct_model()
|
27 |
-
|
28 |
-
def _construct_model(self):
|
29 |
-
initial_layer = nn.Sequential(
|
30 |
-
nn.Conv2d(
|
31 |
-
self.cfg.channels,
|
32 |
-
self.cfg.num_features,
|
33 |
-
kernel_size=7,
|
34 |
-
stride=1,
|
35 |
-
padding=3,
|
36 |
-
padding_mode="reflect",
|
37 |
-
),
|
38 |
-
nn.ReLU(inplace=True),
|
39 |
-
)
|
40 |
-
|
41 |
-
down_blocks = nn.Sequential(
|
42 |
-
ConvBlock(
|
43 |
-
self.cfg.num_features,
|
44 |
-
self.cfg.num_features * 2,
|
45 |
-
kernel_size=3,
|
46 |
-
stride=2,
|
47 |
-
padding=1,
|
48 |
-
),
|
49 |
-
ConvBlock(
|
50 |
-
self.cfg.num_features * 2,
|
51 |
-
self.cfg.num_features * 4,
|
52 |
-
kernel_size=3,
|
53 |
-
stride=2,
|
54 |
-
padding=1,
|
55 |
-
),
|
56 |
-
)
|
57 |
-
|
58 |
-
residual_blocks = nn.Sequential(
|
59 |
-
*[
|
60 |
-
ResidualBlock(self.cfg.num_features * 4)
|
61 |
-
for _ in range(self.cfg.num_residuals)
|
62 |
-
]
|
63 |
-
)
|
64 |
-
|
65 |
-
up_blocks = nn.Sequential(
|
66 |
-
ConvBlock(
|
67 |
-
self.cfg.num_features * 4,
|
68 |
-
self.cfg.num_features * 2,
|
69 |
-
down=False,
|
70 |
-
kernel_size=3,
|
71 |
-
stride=2,
|
72 |
-
padding=1,
|
73 |
-
output_padding=1,
|
74 |
-
),
|
75 |
-
ConvBlock(
|
76 |
-
self.cfg.num_features * 2,
|
77 |
-
self.cfg.num_features,
|
78 |
-
down=False,
|
79 |
-
kernel_size=3,
|
80 |
-
stride=2,
|
81 |
-
padding=1,
|
82 |
-
output_padding=1,
|
83 |
-
),
|
84 |
-
)
|
85 |
-
|
86 |
-
last_layer = nn.Conv2d(
|
87 |
-
self.cfg.num_features,
|
88 |
-
self.cfg.channels,
|
89 |
-
kernel_size=7,
|
90 |
-
stride=1,
|
91 |
-
padding=3,
|
92 |
-
padding_mode="reflect",
|
93 |
-
)
|
94 |
-
|
95 |
-
return nn.Sequential(
|
96 |
-
initial_layer, down_blocks, residual_blocks, up_blocks, last_layer
|
97 |
-
)
|
98 |
-
|
99 |
-
def forward(self, x: Tensor) -> Tensor:
|
100 |
-
return tanh(self.model(x))
|
101 |
-
|
102 |
-
|
103 |
-
class ConvBlock(nn.Module):
|
104 |
-
def __init__(
|
105 |
-
self, in_channels, out_channels, down=True, use_activation=True, **kwargs
|
106 |
-
):
|
107 |
-
super().__init__()
|
108 |
-
self.conv = nn.Sequential(
|
109 |
-
nn.Conv2d(in_channels, out_channels, padding_mode="reflect", **kwargs)
|
110 |
-
if down
|
111 |
-
else nn.ConvTranspose2d(in_channels, out_channels, **kwargs),
|
112 |
-
nn.InstanceNorm2d(out_channels),
|
113 |
-
nn.ReLU(inplace=True) if use_activation else nn.Identity(),
|
114 |
-
)
|
115 |
-
|
116 |
-
def forward(self, x: Tensor) -> Tensor:
|
117 |
-
return self.conv(x)
|
118 |
-
|
119 |
-
|
120 |
-
class ResidualBlock(nn.Module):
|
121 |
-
def __init__(self, channels: int):
|
122 |
-
super().__init__()
|
123 |
-
self.block = nn.Sequential(
|
124 |
-
ConvBlock(channels, channels, kernel_size=3, padding=1),
|
125 |
-
ConvBlock(
|
126 |
-
channels, channels, use_activation=False, kernel_size=3, padding=1
|
127 |
-
),
|
128 |
-
)
|
129 |
-
|
130 |
-
def forward(self, x: Tensor) -> Tensor:
|
131 |
-
return x + self.block(x)
|
132 |
|
|
|
|
|
|
|
133 |
|
134 |
repo_id = "Kiwinicki/sat2map-generator"
|
135 |
generator_path = hf_hub_download(repo_id=repo_id, filename="generator.pth")
|
@@ -145,10 +25,23 @@ generator = Generator(cfg)
|
|
145 |
generator.load_state_dict(torch.load(generator_path))
|
146 |
generator.eval()
|
147 |
|
|
|
|
|
|
|
148 |
|
|
|
|
|
|
|
|
|
|
|
149 |
|
150 |
-
def
|
151 |
-
|
|
|
|
|
|
|
|
|
|
|
152 |
|
153 |
-
iface = gr.Interface(fn=
|
154 |
-
iface.launch()
|
|
|
6 |
import torch
|
7 |
import json
|
8 |
from omegaconf import OmegaConf
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
+
import sys
|
11 |
+
sys.path.append(os.path.dirname(model_path))
|
12 |
+
from model import Generator
|
13 |
|
14 |
repo_id = "Kiwinicki/sat2map-generator"
|
15 |
generator_path = hf_hub_download(repo_id=repo_id, filename="generator.pth")
|
|
|
25 |
generator.load_state_dict(torch.load(generator_path))
|
26 |
generator.eval()
|
27 |
|
28 |
+
from PIL import Image
|
29 |
+
import torchvision.transforms as transforms
|
30 |
+
|
31 |
|
32 |
+
transform = transforms.Compose([
|
33 |
+
transforms.Resize((256, 256)),
|
34 |
+
transforms.ToTensor(),
|
35 |
+
transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
|
36 |
+
])
|
37 |
|
38 |
+
def process_image(image):
|
39 |
+
image_tensor = transform(image).unsqueeze(0)
|
40 |
+
with torch.no_grad():
|
41 |
+
output_tensor = generator(image_tensor)
|
42 |
+
output_image = output_tensor.squeeze(0)
|
43 |
+
output_image = transforms.ToPILImage()(output_image)
|
44 |
+
return output_image
|
45 |
|
46 |
+
iface = gr.Interface(fn=process_image, inputs="image", outputs="image", title="Image Generator")
|
47 |
+
iface.launch()
|