Spaces:
Sleeping
Sleeping
Commit
·
d09b148
1
Parent(s):
e430625
Update app.py to load the GAN model
Browse files
app.py
CHANGED
@@ -1,19 +1,63 @@
|
|
1 |
import torch
|
|
|
2 |
|
3 |
from PIL import Image
|
4 |
|
5 |
import numpy as np
|
6 |
import gradio as gr
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
MODEL_PATH = './gan_mnist_generator_20.pt'
|
9 |
-
model =
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
-
def run_generative_model(use_seed="False", seed=
|
12 |
if use_seed == "True":
|
13 |
torch.random.manual_seed(seed)
|
14 |
|
15 |
# Run generator model
|
16 |
-
noise = torch.randn(1,
|
17 |
with torch.no_grad():
|
18 |
im = model(noise).detach().cpu()
|
19 |
|
|
|
1 |
import torch
|
2 |
+
import torch.nn as nn
|
3 |
|
4 |
from PIL import Image
|
5 |
|
6 |
import numpy as np
|
7 |
import gradio as gr
|
8 |
|
9 |
+
|
10 |
+
class DCGAN_Generator(nn.Module):
|
11 |
+
def __init__(self):
|
12 |
+
super(DCGAN_Generator, self).__init__()
|
13 |
+
|
14 |
+
self.conv1 = nn.ConvTranspose2d(100, 256, 5)
|
15 |
+
self.bn1 = nn.BatchNorm2d(256)
|
16 |
+
self.relu1 = nn.LeakyReLU(negative_slope=0.2)
|
17 |
+
|
18 |
+
self.conv2 = nn.ConvTranspose2d(256, 256, 5)
|
19 |
+
self.bn2 = nn.BatchNorm2d(256)
|
20 |
+
self.relu2 = nn.LeakyReLU(negative_slope=0.2)
|
21 |
+
|
22 |
+
self.conv3 = nn.ConvTranspose2d(256, 128, 4)
|
23 |
+
self.bn3 = nn.BatchNorm2d(128)
|
24 |
+
self.relu3 = nn.LeakyReLU(negative_slope=0.2)
|
25 |
+
|
26 |
+
self.conv4 = nn.ConvTranspose2d(128, 64, 2, 2)
|
27 |
+
self.bn4 = nn.BatchNorm2d(64)
|
28 |
+
self.relu4 = nn.LeakyReLU(negative_slope=0.2)
|
29 |
+
|
30 |
+
self.conv5 = nn.ConvTranspose2d(64, 32, 3)
|
31 |
+
self.bn5 = nn.BatchNorm2d(32)
|
32 |
+
self.relu5 = nn.LeakyReLU(negative_slope=0.2)
|
33 |
+
|
34 |
+
self.conv6 = nn.ConvTranspose2d(32, 1, 3)
|
35 |
+
self.tanh1 = nn.Tanh()
|
36 |
+
|
37 |
+
def forward(self, x):
|
38 |
+
x = self.relu1(self.bn1(self.conv1(x)))
|
39 |
+
x = self.relu2(self.bn2(self.conv2(x)))
|
40 |
+
x = self.relu3(self.bn3(self.conv3(x)))
|
41 |
+
x = self.relu4(self.bn4(self.conv4(x)))
|
42 |
+
x = self.relu5(self.bn5(self.conv5(x)))
|
43 |
+
|
44 |
+
return self.tanh1(self.conv6(x))
|
45 |
+
|
46 |
MODEL_PATH = './gan_mnist_generator_20.pt'
|
47 |
+
model = DCGAN_Generator()
|
48 |
+
model.load_state_dict(
|
49 |
+
torch.load(
|
50 |
+
MODEL_PATH,
|
51 |
+
map_location=torch.device('cpu')
|
52 |
+
)
|
53 |
+
)
|
54 |
|
55 |
+
def run_generative_model(use_seed="False", seed=42):
|
56 |
if use_seed == "True":
|
57 |
torch.random.manual_seed(seed)
|
58 |
|
59 |
# Run generator model
|
60 |
+
noise = torch.randn(1, 100, 1, 1)
|
61 |
with torch.no_grad():
|
62 |
im = model(noise).detach().cpu()
|
63 |
|