add upsampler
Browse files
app.py
CHANGED
@@ -2,6 +2,7 @@ import os
|
|
2 |
|
3 |
import gradio as gr
|
4 |
import torch
|
|
|
5 |
from basicsr.archs.srvgg_arch import SRVGGNetCompact
|
6 |
from gfpgan.utils import GFPGANer
|
7 |
from huggingface_hub import hf_hub_download
|
@@ -12,46 +13,115 @@ GFPGAN_REPO_ID = 'leonelhs/gfpgan'
|
|
12 |
|
13 |
os.system("pip freeze")
|
14 |
|
15 |
-
# background enhancer with RealESRGAN
|
16 |
-
model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu')
|
17 |
-
model_path = hf_hub_download(repo_id=REALESRGAN_REPO_ID, filename='realesr-general-x4v3.pth')
|
18 |
-
half = True if torch.cuda.is_available() else False
|
19 |
-
upsampler = RealESRGANer(scale=4, model_path=model_path, model=model, tile=0, tile_pad=10, pre_pad=0, half=half)
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
-
|
|
|
23 |
return hf_hub_download(repo_id=GFPGAN_REPO_ID, filename=file)
|
24 |
|
25 |
|
26 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
scale = int(scale)
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
face_enhancer =
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
model_path=path, upscale=scale, arch='RestoreFormer', channel_multiplier=2, bg_upsampler=upsampler)
|
46 |
-
|
47 |
-
_, _, output = face_enhancer.enhance(image, has_aligned=False, only_center_face=False, paste_back=True)
|
48 |
-
|
49 |
-
return output
|
50 |
-
|
51 |
-
|
52 |
-
title = "GFPGAN"
|
53 |
description = r"""
|
54 |
-
<b>Practical
|
55 |
"""
|
56 |
article = r"""
|
57 |
<center><span>[email protected] or [email protected]</span></center>
|
@@ -62,13 +132,31 @@ article = r"""
|
|
62 |
demo = gr.Interface(
|
63 |
predict, [
|
64 |
gr.Image(type="numpy", label="Input"),
|
65 |
-
gr.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
gr.Dropdown(["1", "2", "3", "4"], value="2", label="Rescaling factor")
|
67 |
], [
|
68 |
-
gr.Image(type="numpy", label="Output", interactive=False)
|
|
|
69 |
],
|
70 |
title=title,
|
71 |
description=description,
|
72 |
article=article)
|
73 |
|
74 |
-
demo.queue().launch()
|
|
|
2 |
|
3 |
import gradio as gr
|
4 |
import torch
|
5 |
+
from basicsr.archs.rrdbnet_arch import RRDBNet
|
6 |
from basicsr.archs.srvgg_arch import SRVGGNetCompact
|
7 |
from gfpgan.utils import GFPGANer
|
8 |
from huggingface_hub import hf_hub_download
|
|
|
13 |
|
14 |
os.system("pip freeze")
|
15 |
|
|
|
|
|
|
|
|
|
|
|
16 |
|
17 |
+
def showGPU():
|
18 |
+
if torch.cuda.is_available():
|
19 |
+
devices = torch.cuda.device_count()
|
20 |
+
current = torch.cuda.current_device()
|
21 |
+
return f"Running on GPU:{current} of {devices} total devices"
|
22 |
+
return "Running on CPU"
|
23 |
|
24 |
+
|
25 |
+
def download_model_gfpgan(file):
|
26 |
return hf_hub_download(repo_id=GFPGAN_REPO_ID, filename=file)
|
27 |
|
28 |
|
29 |
+
def download_model_realesrgan(file):
|
30 |
+
return hf_hub_download(repo_id=REALESRGAN_REPO_ID, filename=file)
|
31 |
+
|
32 |
+
|
33 |
+
def select_upsampler(version, netscale=4):
|
34 |
+
model = None
|
35 |
+
dni_weight = None
|
36 |
+
|
37 |
+
version = version + ".pth"
|
38 |
+
model_path = download_model_realesrgan(version)
|
39 |
+
|
40 |
+
if version == 'RealESRGAN_x4plus.pth': # x4 RRDBNet model
|
41 |
+
model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
|
42 |
+
|
43 |
+
if version == 'RealESRNet_x4plus.pth': # x4 RRDBNet model
|
44 |
+
model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
|
45 |
+
|
46 |
+
if version == 'AI-Forever_x4plus.pth': # x4 RRDBNet model
|
47 |
+
model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
|
48 |
+
|
49 |
+
if version == 'RealESRGAN_x4plus_anime_6B.pth': # x4 RRDBNet model with 6 blocks
|
50 |
+
model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=6, num_grow_ch=32, scale=4)
|
51 |
+
|
52 |
+
if version == 'RealESRGAN_x2plus.pth': # x2 RRDBNet model
|
53 |
+
model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=2)
|
54 |
+
netscale = 2 # This is
|
55 |
+
|
56 |
+
if version == 'AI-Forever_x2plus.pth': # x2 RRDBNet model
|
57 |
+
model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=2)
|
58 |
+
netscale = 2 # This is
|
59 |
+
|
60 |
+
if version == 'realesr-animevideov3.pth': # x4 VGG-style model (XS size)
|
61 |
+
model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=16, upscale=4, act_type='prelu')
|
62 |
+
|
63 |
+
if version == 'realesr-general-x4v3.pth': # x4 VGG-style model (S size)
|
64 |
+
model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu')
|
65 |
+
model_path = [
|
66 |
+
download_model_realesrgan("realesr-general-x4v3.pth"),
|
67 |
+
download_model_realesrgan("realesr-general-wdn-x4v3.pth")
|
68 |
+
]
|
69 |
+
dni_weight = [0.2, 0.8]
|
70 |
+
|
71 |
+
half = True if torch.cuda.is_available() else False
|
72 |
+
|
73 |
+
return RealESRGANer(
|
74 |
+
scale=netscale,
|
75 |
+
model_path=model_path,
|
76 |
+
dni_weight=dni_weight,
|
77 |
+
model=model,
|
78 |
+
tile=0,
|
79 |
+
tile_pad=10,
|
80 |
+
pre_pad=0,
|
81 |
+
half=half,
|
82 |
+
gpu_id=0)
|
83 |
+
|
84 |
+
|
85 |
+
def select_face_enhancer(version, scale, upsampler):
|
86 |
+
if 'v1.2' in version:
|
87 |
+
model_path = download_model_gfpgan('GFPGANv1.2.pth')
|
88 |
+
return GFPGANer(
|
89 |
+
model_path=model_path, upscale=scale, arch='clean', channel_multiplier=2, bg_upsampler=upsampler)
|
90 |
+
elif 'v1.3' in version:
|
91 |
+
model_path = download_model_gfpgan('GFPGANv1.3.pth')
|
92 |
+
return GFPGANer(
|
93 |
+
model_path=model_path, upscale=scale, arch='clean', channel_multiplier=2, bg_upsampler=upsampler)
|
94 |
+
elif 'v1.4' in version:
|
95 |
+
model_path = download_model_gfpgan('GFPGANv1.4.pth')
|
96 |
+
return GFPGANer(
|
97 |
+
model_path=model_path, upscale=scale, arch='clean', channel_multiplier=2, bg_upsampler=upsampler)
|
98 |
+
elif 'RestoreFormer' in version:
|
99 |
+
model_path = download_model_gfpgan('RestoreFormer.pth')
|
100 |
+
return GFPGANer(
|
101 |
+
model_path=model_path, upscale=scale, arch='RestoreFormer', channel_multiplier=2, bg_upsampler=upsampler)
|
102 |
+
|
103 |
+
|
104 |
+
def predict(image, version_upsampler, version_enhancer, scale):
|
105 |
scale = int(scale)
|
106 |
+
|
107 |
+
upsampler = select_upsampler(version_upsampler)
|
108 |
+
|
109 |
+
if "No additional" not in version_enhancer:
|
110 |
+
face_enhancer = select_face_enhancer(version_enhancer, scale, upsampler)
|
111 |
+
_, _, output = face_enhancer.enhance(image, has_aligned=False, only_center_face=False, paste_back=True)
|
112 |
+
else:
|
113 |
+
output, _ = upsampler.enhance(image, outscale=scale)
|
114 |
+
|
115 |
+
log = f"General enhance version: {version_upsampler}\n " \
|
116 |
+
f"Face enhance version: {version_enhancer} \n " \
|
117 |
+
f"Scale:{scale} \n {showGPU()}"
|
118 |
+
|
119 |
+
return output, log
|
120 |
+
|
121 |
+
|
122 |
+
title = "Super Face"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
description = r"""
|
124 |
+
<b>Practical Image Restoration Algorithm based on Real-ESRGAN, GFPGAN</b>
|
125 |
"""
|
126 |
article = r"""
|
127 |
<center><span>[email protected] or [email protected]</span></center>
|
|
|
132 |
demo = gr.Interface(
|
133 |
predict, [
|
134 |
gr.Image(type="numpy", label="Input"),
|
135 |
+
gr.Dropdown([
|
136 |
+
'RealESRGAN_x2plus',
|
137 |
+
'RealESRGAN_x4plus',
|
138 |
+
'RealESRNet_x4plus',
|
139 |
+
'AI-Forever_x2plus',
|
140 |
+
'AI-Forever_x4plus',
|
141 |
+
'RealESRGAN_x4plus_anime_6B',
|
142 |
+
'realesr-animevideov3',
|
143 |
+
'realesr-general-x4v3'],
|
144 |
+
type="value", value='RealESRGAN_x4plus', label='General restoration algorithm', info="version"),
|
145 |
+
gr.Dropdown([
|
146 |
+
'No additional face process',
|
147 |
+
'GFPGANv1.2',
|
148 |
+
'GFPGANv1.3',
|
149 |
+
'GFPGANv1.4',
|
150 |
+
'RestoreFormer'],
|
151 |
+
type="value", value='No additional face process', label='Special face restoration algorithm',
|
152 |
+
info="version"),
|
153 |
gr.Dropdown(["1", "2", "3", "4"], value="2", label="Rescaling factor")
|
154 |
], [
|
155 |
+
gr.Image(type="numpy", label="Output", interactive=False),
|
156 |
+
gr.Textbox(label="log info")
|
157 |
],
|
158 |
title=title,
|
159 |
description=description,
|
160 |
article=article)
|
161 |
|
162 |
+
demo.queue().launch(share=True, debug=True)
|