Update app.py
Browse files
app.py
CHANGED
@@ -1,9 +1,57 @@
|
|
1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
|
|
|
|
|
|
3 |
|
4 |
-
def greet(name):
|
5 |
-
return "Hello " + name + "!!"
|
6 |
|
7 |
-
demo = gr.Interface(fn=greet, inputs="text", outputs="text")
|
8 |
|
9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from torch import nn
|
2 |
+
|
3 |
+
class Generator(nn.Module):
|
4 |
+
# Refer to the link below for explanations about nc, nz, and ngf
|
5 |
+
# https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html#inputs
|
6 |
+
def __init__(self, nc=4, nz=100, ngf=64):
|
7 |
+
super(Generator, self).__init__()
|
8 |
+
self.network = nn.Sequential(
|
9 |
+
nn.ConvTranspose2d(nz, ngf * 4, 3, 1, 0, bias=False),
|
10 |
+
nn.BatchNorm2d(ngf * 4),
|
11 |
+
nn.ReLU(True),
|
12 |
+
nn.ConvTranspose2d(ngf * 4, ngf * 2, 3, 2, 1, bias=False),
|
13 |
+
nn.BatchNorm2d(ngf * 2),
|
14 |
+
nn.ReLU(True),
|
15 |
+
nn.ConvTranspose2d(ngf * 2, ngf, 4, 2, 0, bias=False),
|
16 |
+
nn.BatchNorm2d(ngf),
|
17 |
+
nn.ReLU(True),
|
18 |
+
nn.ConvTranspose2d(ngf, nc, 4, 2, 1, bias=False),
|
19 |
+
nn.Tanh(),
|
20 |
+
)
|
21 |
+
|
22 |
+
def forward(self, input):
|
23 |
+
output = self.network(input)
|
24 |
+
return output
|
25 |
+
|
26 |
+
|
27 |
+
|
28 |
+
from huggingface_hub import hf_hub_download
|
29 |
+
import torch
|
30 |
|
31 |
+
model = Generator()
|
32 |
+
weights_path = hf_hub_download('nateraw/cryptopunks-gan', 'generator.pth')
|
33 |
+
model.load_state_dict(torch.load(weights_path, map_location=torch.device('cpu'))) # Use 'cuda' if you have a GPU available
|
34 |
|
|
|
|
|
35 |
|
|
|
36 |
|
37 |
+
from torchvision.utils import save_image
|
38 |
+
|
39 |
+
def predict(seed):
|
40 |
+
num_punks = 4
|
41 |
+
torch.manual_seed(seed)
|
42 |
+
z = torch.randn(num_punks, 100, 1, 1)
|
43 |
+
punks = model(z)
|
44 |
+
save_image(punks, "punks.png", normalize=True)
|
45 |
+
return 'punks.png'
|
46 |
+
|
47 |
+
|
48 |
+
|
49 |
+
import gradio as gr
|
50 |
+
|
51 |
+
gr.Interface(
|
52 |
+
predict,
|
53 |
+
inputs=[
|
54 |
+
gr.Slider(0, 1000, label='Seed', default=42),
|
55 |
+
],
|
56 |
+
outputs="image",
|
57 |
+
).launch()
|