Add files
Browse files- .gitmodules +3 -0
- StyleSwin +1 -0
- app.py +135 -0
- packages.txt +1 -0
- requirements.txt +5 -0
.gitmodules
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
[submodule "StyleSwin"]
|
2 |
+
path = StyleSwin
|
3 |
+
url = https://github.com/microsoft/StyleSwin
|
StyleSwin
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
Subproject commit 52c23dcfa39a5da75f02892cb775fe8f424be6ec
|
app.py
ADDED
@@ -0,0 +1,135 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
|
3 |
+
from __future__ import annotations
|
4 |
+
|
5 |
+
import argparse
|
6 |
+
import functools
|
7 |
+
import os
|
8 |
+
import sys
|
9 |
+
|
10 |
+
sys.path.insert(0, 'StyleSwin')
|
11 |
+
|
12 |
+
import gradio as gr
|
13 |
+
import huggingface_hub
|
14 |
+
import numpy as np
|
15 |
+
import PIL.Image
|
16 |
+
import torch
|
17 |
+
import torch.nn as nn
|
18 |
+
from models.generator import Generator
|
19 |
+
|
20 |
+
TOKEN = os.environ['TOKEN']
|
21 |
+
|
22 |
+
MODEL_REPO = 'hysts/StyleSwin'
|
23 |
+
MODEL_NAMES = [
|
24 |
+
'CelebAHQ_256',
|
25 |
+
'FFHQ_256',
|
26 |
+
'LSUNChurch_256',
|
27 |
+
'CelebAHQ_1024',
|
28 |
+
'FFHQ_1024',
|
29 |
+
]
|
30 |
+
|
31 |
+
|
32 |
+
def parse_args() -> argparse.Namespace:
|
33 |
+
parser = argparse.ArgumentParser()
|
34 |
+
parser.add_argument('--device', type=str, default='cpu')
|
35 |
+
parser.add_argument('--theme', type=str)
|
36 |
+
parser.add_argument('--live', action='store_true')
|
37 |
+
parser.add_argument('--share', action='store_true')
|
38 |
+
parser.add_argument('--port', type=int)
|
39 |
+
parser.add_argument('--disable-queue',
|
40 |
+
dest='enable_queue',
|
41 |
+
action='store_false')
|
42 |
+
parser.add_argument('--allow-flagging', type=str, default='never')
|
43 |
+
parser.add_argument('--allow-screenshot', action='store_true')
|
44 |
+
return parser.parse_args()
|
45 |
+
|
46 |
+
|
47 |
+
def load_model(model_name: str, device: torch.device) -> nn.Module:
|
48 |
+
size = int(model_name.split('_')[1])
|
49 |
+
channel_multiplier = 1 if size == 1024 else 2
|
50 |
+
model = Generator(size,
|
51 |
+
style_dim=512,
|
52 |
+
n_mlp=8,
|
53 |
+
channel_multiplier=channel_multiplier)
|
54 |
+
ckpt_path = huggingface_hub.hf_hub_download(MODEL_REPO,
|
55 |
+
f'models/{model_name}.pt',
|
56 |
+
use_auth_token=TOKEN)
|
57 |
+
ckpt = torch.load(ckpt_path)
|
58 |
+
model.load_state_dict(ckpt['g_ema'])
|
59 |
+
model.to(device)
|
60 |
+
model.eval()
|
61 |
+
return model
|
62 |
+
|
63 |
+
|
64 |
+
def generate_z(seed: int, device: torch.device) -> torch.Tensor:
|
65 |
+
return torch.from_numpy(np.random.RandomState(seed).randn(
|
66 |
+
1, 512)).to(device).float()
|
67 |
+
|
68 |
+
|
69 |
+
def postprocess(tensors: torch.Tensor) -> torch.Tensor:
|
70 |
+
assert tensors.dim() == 4
|
71 |
+
tensors = tensors.cpu()
|
72 |
+
std = torch.FloatTensor([0.229, 0.224, 0.225])[None, :, None, None]
|
73 |
+
mean = torch.FloatTensor([0.485, 0.456, 0.406])[None, :, None, None]
|
74 |
+
tensors = tensors * std + mean
|
75 |
+
tensors = (tensors * 255).clamp(0, 255).to(torch.uint8)
|
76 |
+
return tensors
|
77 |
+
|
78 |
+
|
79 |
+
@torch.inference_mode()
|
80 |
+
def generate_image(model_name: str, seed: int, model_dict: dict,
|
81 |
+
device: torch.device) -> PIL.Image.Image:
|
82 |
+
model = model_dict[model_name]
|
83 |
+
seed = int(np.clip(seed, 0, np.iinfo(np.uint32).max))
|
84 |
+
z = generate_z(seed, device)
|
85 |
+
out, _ = model(z)
|
86 |
+
out = postprocess(out)
|
87 |
+
out = out.numpy()[0].transpose(1, 2, 0)
|
88 |
+
return PIL.Image.fromarray(out, 'RGB')
|
89 |
+
|
90 |
+
|
91 |
+
def main():
|
92 |
+
gr.close_all()
|
93 |
+
|
94 |
+
args = parse_args()
|
95 |
+
device = torch.device(args.device)
|
96 |
+
|
97 |
+
model_dict = {name: load_model(name, device) for name in MODEL_NAMES}
|
98 |
+
|
99 |
+
func = functools.partial(generate_image,
|
100 |
+
model_dict=model_dict,
|
101 |
+
device=device)
|
102 |
+
func = functools.update_wrapper(func, generate_image)
|
103 |
+
|
104 |
+
repo_url = 'https://github.com/microsoft/StyleSwin'
|
105 |
+
title = 'microsoft/StyleSwin'
|
106 |
+
description = f'A demo for {repo_url}'
|
107 |
+
article = None
|
108 |
+
|
109 |
+
gr.Interface(
|
110 |
+
func,
|
111 |
+
[
|
112 |
+
gr.inputs.Radio(MODEL_NAMES,
|
113 |
+
type='value',
|
114 |
+
default='FFHQ_256',
|
115 |
+
label='model',
|
116 |
+
optional=False),
|
117 |
+
gr.inputs.Slider(0, 2147483647, step=1, default=0, label='Seed'),
|
118 |
+
],
|
119 |
+
gr.outputs.Image(type='pil', label='Output'),
|
120 |
+
theme=args.theme,
|
121 |
+
title=title,
|
122 |
+
description=description,
|
123 |
+
article=article,
|
124 |
+
allow_screenshot=args.allow_screenshot,
|
125 |
+
allow_flagging=args.allow_flagging,
|
126 |
+
live=args.live,
|
127 |
+
).launch(
|
128 |
+
enable_queue=args.enable_queue,
|
129 |
+
server_port=args.port,
|
130 |
+
share=args.share,
|
131 |
+
)
|
132 |
+
|
133 |
+
|
134 |
+
if __name__ == '__main__':
|
135 |
+
main()
|
packages.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
ninja-build
|
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
numpy==1.22.3
|
2 |
+
Pillow==9.0.1
|
3 |
+
timm==0.5.4
|
4 |
+
torch==1.11.0
|
5 |
+
torchvision==0.12.0
|