-
Notifications
You must be signed in to change notification settings - Fork 49
/
app.py
209 lines (187 loc) · 8.33 KB
/
app.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
#!/usr/bin/env python
# -*- coding:utf-8 -*-
# Power by Zongsheng Yue 2023-08-15 09:39:58
import argparse
import gradio as gr
from pathlib import Path
from omegaconf import OmegaConf
from sampler import ResShiftSampler
from utils import util_image
from basicsr.utils.download_util import load_file_from_url
_STEP = {
'v1': 15,
'v2': 15,
'v3': 4,
'bicsr': 4,
'inpaint_imagenet': 4,
'inpaint_face': 4,
'faceir': 4,
'deblur': 4,
}
_LINK = {
'vqgan': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/autoencoder_vq_f4.pth',
'vqgan_face256': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/celeba256_vq_f4_dim3_face.pth',
'vqgan_face512': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/ffhq512_vq_f8_dim8_face.pth',
'v1': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_realsrx4_s15_v1.pth',
'v2': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_realsrx4_s15_v2.pth',
'v3': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_realsrx4_s4_v3.pth',
'bicsr': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_bicsrx4_s4.pth',
'inpaint_imagenet': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_inpainting_imagenet_s4.pth',
'inpaint_face': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_inpainting_face_s4.pth',
'faceir': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_faceir_s4.pth',
'deblur': 'https://github.com/zsyOAOA/ResShift/releases/download/v2.0/resshift_deblur_s4.pth',
}
def get_configs(task='realsr', version='v3', scale=4):
ckpt_dir = Path('./weights')
if not ckpt_dir.exists():
ckpt_dir.mkdir()
if task == 'realsr':
if version in ['v1', 'v2']:
configs = OmegaConf.load('./configs/realsr_swinunet_realesrgan256.yaml')
elif version == 'v3':
configs = OmegaConf.load('./configs/realsr_swinunet_realesrgan256_journal.yaml')
else:
raise ValueError(f"Unexpected version type: {version}")
assert scale == 4, 'We only support the 4x super-resolution now!'
ckpt_url = _LINK[version]
ckpt_path = ckpt_dir / f'resshift_{task}x{scale}_s{_STEP[version]}_{version}.pth'
vqgan_url = _LINK['vqgan']
vqgan_path = ckpt_dir / f'autoencoder_vq_f4.pth'
elif task == 'bicsr':
configs = OmegaConf.load('./configs/bicx4_swinunet_lpips.yaml')
assert scale == 4, 'We only support the 4x super-resolution now!'
ckpt_url = _LINK[task]
ckpt_path = ckpt_dir / f'resshift_{task}x{scale}_s{_STEP[task]}.pth'
vqgan_url = _LINK['vqgan']
vqgan_path = ckpt_dir / f'autoencoder_vq_f4.pth'
# elif task == 'inpaint_imagenet':
# configs = OmegaConf.load('./configs/inpaint_lama256_imagenet.yaml')
# assert scale == 1, 'Please set scale equals 1 for image inpainting!'
# ckpt_url = _LINK[task]
# ckpt_path = ckpt_dir / f'resshift_{task}_s{_STEP[task]}.pth'
# vqgan_url = _LINK['vqgan']
# vqgan_path = ckpt_dir / f'autoencoder_vq_f4.pth'
# elif task == 'inpaint_face':
# configs = OmegaConf.load('./configs/inpaint_lama256_face.yaml')
# assert scale == 1, 'Please set scale equals 1 for image inpainting!'
# ckpt_url = _LINK[task]
# ckpt_path = ckpt_dir / f'resshift_{task}_s{_STEP[task]}.pth'
# vqgan_url = _LINK['vqgan_face256']
# vqgan_path = ckpt_dir / f'celeba256_vq_f4_dim3_face.pth'
# elif task == 'faceir':
# configs = OmegaConf.load('./configs/faceir_gfpgan512_lpips.yaml')
# assert scale == 1, 'Please set scale equals 1 for face restoration!'
# ckpt_url = _LINK[task]
# ckpt_path = ckpt_dir / f'resshift_{task}_s{_STEP[task]}.pth'
# vqgan_url = _LINK['vqgan_face512']
# vqgan_path = ckpt_dir / f'ffhq512_vq_f8_dim8_face.pth'
# elif args.task == 'deblur':
# configs = OmegaConf.load('./configs/deblur_gopro256.yaml')
# assert args.scale == 1, 'Please set scale equals 1 for deblurring!'
# ckpt_url = _LINK[args.task]
# ckpt_path = ckpt_dir / f'resshift_{args.task}_s{_STEP[args.task]}.pth'
# vqgan_url = _LINK['vqgan']
# vqgan_path = ckpt_dir / f'autoencoder_vq_f4.pth'
else:
raise TypeError(f"Unexpected task type: {task}!")
# prepare the checkpoint
if not ckpt_path.exists():
load_file_from_url(
url=ckpt_url,
model_dir=ckpt_dir,
progress=True,
file_name=ckpt_path.name,
)
if not vqgan_path.exists():
load_file_from_url(
url=vqgan_url,
model_dir=ckpt_dir,
progress=True,
file_name=vqgan_path.name,
)
configs.model.ckpt_path = str(ckpt_path)
configs.diffusion.params.sf = scale
configs.autoencoder.ckpt_path = str(vqgan_path)
return configs
def predict(in_path, task='realsrx4', seed=12345, scale=4, version='v3'):
configs = get_configs(task, version, scale)
resshift_sampler = ResShiftSampler(
configs,
sf=scale,
chop_size=256,
chop_stride=224,
chop_bs=1,
use_amp=True,
seed=seed,
padding_offset=configs.model.params.get('lq_size', 64),
)
out_dir = Path('restored_output')
if not out_dir.exists():
out_dir.mkdir()
resshift_sampler.inference(
in_path,
out_dir,
mask_path=None,
bs=1,
noise_repeat=False
)
out_path = out_dir / f"{Path(in_path).stem}.png"
assert out_path.exists(), 'Super-resolution failed!'
im_sr = util_image.imread(out_path, chn="rgb", dtype="uint8")
return im_sr, str(out_path)
title = "ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting"
description = r"""
<b>Official Gradio demo</b> for <a href='https://github.com/zsyOAOA/ResShift' target='_blank'><b>ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting</b></a>.<br>
🔥 ResShift is an efficient diffusion model designed for image super-resolution or restoration.<br>
"""
article = r"""
If ResShift is helpful for your work, please help to ⭐ the <a href='https://github.com/zsyOAOA/ResShift' target='_blank'>Github Repo</a>. Thanks!
[](https://github.com/zsyOAOA/ResShift)
---
If our work is useful for your research, please consider citing:
```bibtex
@inproceedings{yue2023resshift,
title={ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting},
author={Yue, Zongsheng and Wang, Jianyi and Loy, Chen Change},
booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
year={2023},
volume = {36},
pages = {13294--13307},
}
```
📋 **License**
This project is licensed under <a rel="license" href="https://github.com/zsyOAOA/ResShift/blob/master/LICENSE">S-Lab License 1.0</a>.
Redistribution and use for non-commercial purposes should follow this license.
📧 **Contact**
If you have any questions, please feel free to contact me via <b>[email protected]</b>.
"""
demo = gr.Interface(
fn=predict,
inputs=[
gr.Image(type="filepath", label="Input: Low Quality Image"),
gr.Dropdown(
choices=["realsr", "bicsr"],
value="realsr",
label="Task",
),
gr.Number(value=12345, precision=0, label="Ranom seed")
],
outputs=[
gr.Image(type="numpy", label="Output: High Quality Image"),
gr.outputs.File(label="Download the output")
],
title=title,
description=description,
article=article,
examples=[
['./testdata/RealSet65/0030.jpg', "realsr", 12345],
['./testdata/RealSet65/dog2.png', "realsr", 12345],
['./testdata/RealSet65/bears.jpg', "realsr", 12345],
['./testdata/RealSet65/oldphoto6.png', "realsr", 12345],
['./testdata/Bicubicx4/lq_matlab/ILSVRC2012_val_00000067.png', "bicsr", 12345],
['./testdata/Bicubicx4/lq_matlab/ILSVRC2012_val_00016898.png', "bicsr", 12345],
]
)
demo.queue(concurrency_count=4)
demo.launch(share=True)