Spaces:
Sleeping
Sleeping
init
Browse files- app.py +46 -0
- images/18.encoded.png +0 -0
- images/20.encoded.png +0 -0
- images/21.encoded.png +0 -0
app.py
ADDED
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
|
4 |
+
import torch
|
5 |
+
from diffusers import AutoencoderTiny
|
6 |
+
from torchvision.transforms.functional import to_pil_image, center_crop, resize, to_tensor
|
7 |
+
|
8 |
+
device = 'cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu'
|
9 |
+
d_type = torch.float32 if device == 'mps' else torch.float16
|
10 |
+
|
11 |
+
model_id = "madebyollin/taesd"
|
12 |
+
vae = AutoencoderTiny.from_pretrained(model_id, safetensors=True, torch_dtype=d_type).to(device)
|
13 |
+
|
14 |
+
|
15 |
+
@torch.no_grad()
|
16 |
+
def decode(image):
|
17 |
+
t = to_tensor(image).unsqueeze(0).to(device, dtype=d_type)
|
18 |
+
unscaled_t = vae.unscale_latents(t)
|
19 |
+
reconstructed = vae.decoder(unscaled_t).clamp(0, 1)
|
20 |
+
return to_pil_image(reconstructed[0])
|
21 |
+
|
22 |
+
|
23 |
+
astronaut = os.path.join(os.path.dirname(__file__), "images/21.encoded.png")
|
24 |
+
|
25 |
+
|
26 |
+
def app():
|
27 |
+
return gr.Interface(decode,
|
28 |
+
gr.Image(type="pil",
|
29 |
+
image_mode="RGBA",
|
30 |
+
mirror_webcam=False,
|
31 |
+
label='64x64',
|
32 |
+
value=astronaut),
|
33 |
+
gr.Image(type="pil",
|
34 |
+
image_mode="RGB",
|
35 |
+
label='512x512',
|
36 |
+
height=256,
|
37 |
+
width=256
|
38 |
+
),
|
39 |
+
examples=[
|
40 |
+
os.path.join(os.path.dirname(__file__), "images/18.encoded.png"),
|
41 |
+
os.path.join(os.path.dirname(__file__), "images/20.encoded.png")
|
42 |
+
])
|
43 |
+
|
44 |
+
|
45 |
+
if __name__ == "__main__":
|
46 |
+
app().launch()
|
images/18.encoded.png
ADDED
images/20.encoded.png
ADDED
images/21.encoded.png
ADDED