Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,166 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
import cv2
|
4 |
+
import shutil
|
5 |
+
import sys
|
6 |
+
from subprocess import call
|
7 |
+
import torch
|
8 |
+
import numpy as np
|
9 |
+
from skimage import color
|
10 |
+
import torchvision.transforms as transforms
|
11 |
+
from PIL import Image
|
12 |
+
import torch
|
13 |
+
import uuid
|
14 |
+
|
15 |
+
#os.system("pip install dlib")
|
16 |
+
os.system('bash setup.sh')
|
17 |
+
|
18 |
+
def lab2rgb(L, AB):
|
19 |
+
"""Convert an Lab tensor image to a RGB numpy output
|
20 |
+
Parameters:
|
21 |
+
L (1-channel tensor array): L channel images (range: [-1, 1], torch tensor array)
|
22 |
+
AB (2-channel tensor array): ab channel images (range: [-1, 1], torch tensor array)
|
23 |
+
Returns:
|
24 |
+
rgb (RGB numpy image): rgb output images (range: [0, 255], numpy array)
|
25 |
+
"""
|
26 |
+
AB2 = AB * 110.0
|
27 |
+
L2 = (L + 1.0) * 50.0
|
28 |
+
Lab = torch.cat([L2, AB2], dim=1)
|
29 |
+
Lab = Lab[0].data.cpu().float().numpy()
|
30 |
+
Lab = np.transpose(Lab.astype(np.float64), (1, 2, 0))
|
31 |
+
rgb = color.lab2rgb(Lab) * 255
|
32 |
+
return rgb
|
33 |
+
|
34 |
+
def get_transform(model_name,params=None, grayscale=False, method=Image.BICUBIC):
|
35 |
+
#params
|
36 |
+
preprocess = 'resize'
|
37 |
+
load_size = 256
|
38 |
+
crop_size = 256
|
39 |
+
transform_list = []
|
40 |
+
if grayscale:
|
41 |
+
transform_list.append(transforms.Grayscale(1))
|
42 |
+
if model_name == "Pix2Pix Unet 256":
|
43 |
+
osize = [load_size, load_size]
|
44 |
+
transform_list.append(transforms.Resize(osize, method))
|
45 |
+
# if 'crop' in preprocess:
|
46 |
+
# if params is None:
|
47 |
+
# transform_list.append(transforms.RandomCrop(crop_size))
|
48 |
+
|
49 |
+
return transforms.Compose(transform_list)
|
50 |
+
|
51 |
+
def inferRestoration(img, model_name):
|
52 |
+
#if model_name == "Pix2Pix":
|
53 |
+
model = torch.hub.load('manhkhanhad/ImageRestorationInfer', 'pix2pixRestoration_unet256')
|
54 |
+
transform_list = [
|
55 |
+
transforms.ToTensor(),
|
56 |
+
transforms.Resize([256,256], Image.BICUBIC),
|
57 |
+
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
|
58 |
+
]
|
59 |
+
transform = transforms.Compose(transform_list)
|
60 |
+
img = transform(img)
|
61 |
+
img = torch.unsqueeze(img, 0)
|
62 |
+
result = model(img)
|
63 |
+
result = result[0].detach()
|
64 |
+
result = (result +1)/2.0
|
65 |
+
|
66 |
+
result = transforms.ToPILImage()(result)
|
67 |
+
return result
|
68 |
+
|
69 |
+
def inferColorization(img):
|
70 |
+
model_name = "Deoldify"
|
71 |
+
model = torch.hub.load('manhkhanhad/ImageRestorationInfer', 'DeOldifyColorization')
|
72 |
+
transform_list = [
|
73 |
+
transforms.ToTensor(),
|
74 |
+
transforms.Normalize((0.5,), (0.5,))
|
75 |
+
]
|
76 |
+
transform = transforms.Compose(transform_list)
|
77 |
+
#a = transforms.ToTensor()(a)
|
78 |
+
img = img.convert('L')
|
79 |
+
img = transform(img)
|
80 |
+
img = torch.unsqueeze(img, 0)
|
81 |
+
result = model(img)
|
82 |
+
|
83 |
+
result = result[0].detach()
|
84 |
+
result = (result +1)/2.0
|
85 |
+
|
86 |
+
#img = transforms.Grayscale(3)(img)
|
87 |
+
#img = transforms.ToTensor()(img)
|
88 |
+
#img = torch.unsqueeze(img, 0)
|
89 |
+
#result = model(img)
|
90 |
+
#result = torch.clip(result, min=0, max=1)
|
91 |
+
image_pil = transforms.ToPILImage()(result)
|
92 |
+
return image_pil
|
93 |
+
|
94 |
+
transform_seq = get_transform(model_name)
|
95 |
+
img = transform_seq(img)
|
96 |
+
# if model_name == "Pix2Pix Unet 256":
|
97 |
+
# img.resize((256,256))
|
98 |
+
img = np.array(img)
|
99 |
+
lab = color.rgb2lab(img).astype(np.float32)
|
100 |
+
lab_t = transforms.ToTensor()(lab)
|
101 |
+
A = lab_t[[0], ...] / 50.0 - 1.0
|
102 |
+
B = lab_t[[1, 2], ...] / 110.0
|
103 |
+
#data = {'A': A, 'B': B, 'A_paths': "", 'B_paths': ""}
|
104 |
+
L = torch.unsqueeze(A, 0)
|
105 |
+
#print(L.shape)
|
106 |
+
ab = model(L)
|
107 |
+
Lab = lab2rgb(L, ab).astype(np.uint8)
|
108 |
+
image_pil = Image.fromarray(Lab)
|
109 |
+
#image_pil.save('test.png')
|
110 |
+
#print(Lab.shape)
|
111 |
+
return image_pil
|
112 |
+
|
113 |
+
def colorizaition(image,model_name):
|
114 |
+
image = Image.fromarray(image)
|
115 |
+
result = inferColorization(image,model_name)
|
116 |
+
return result
|
117 |
+
|
118 |
+
|
119 |
+
def run_cmd(command):
|
120 |
+
try:
|
121 |
+
call(command, shell=True)
|
122 |
+
except KeyboardInterrupt:
|
123 |
+
print("Process interrupted")
|
124 |
+
sys.exit(1)
|
125 |
+
|
126 |
+
def run(image):
|
127 |
+
uid = uuid.uuid4()
|
128 |
+
|
129 |
+
if os.path.isdir(f"Temp{uid}"):
|
130 |
+
shutil.rmtree(f"Temp{uid}")
|
131 |
+
|
132 |
+
os.makedirs(f"Temp{uid}")
|
133 |
+
os.makedirs(f"Temp{uid}/input")
|
134 |
+
print(type(image))
|
135 |
+
cv2.imwrite(f"Temp{uid}/input/input_img.png", image)
|
136 |
+
|
137 |
+
command = ("python run.py --input_folder "
|
138 |
+
+ f"Temp{uid}/input"
|
139 |
+
+ " --output_folder "
|
140 |
+
+ f"Temp{uid}"
|
141 |
+
+ " --GPU "
|
142 |
+
+ "-1"
|
143 |
+
+ " --with_scratch")
|
144 |
+
run_cmd(command)
|
145 |
+
|
146 |
+
result_restoration = Image.open(f"Temp{uid}/final_output/input_img.png")
|
147 |
+
shutil.rmtree(f"Temp{uid}")
|
148 |
+
|
149 |
+
result_colorization = inferColorization(result_restoration)
|
150 |
+
|
151 |
+
return result_colorization
|
152 |
+
def load_im(url):
|
153 |
+
return url
|
154 |
+
|
155 |
+
|
156 |
+
with gr.Blocks() as app:
|
157 |
+
im = gr.Image(label="Input Image")
|
158 |
+
with gr.Row():
|
159 |
+
im_u = gr.Textbox()
|
160 |
+
lim_btn=gr.Button("Load")
|
161 |
+
im_btn=gr.Button(label="Restore")
|
162 |
+
out_im = gr.Image(label="Restored Image")
|
163 |
+
|
164 |
+
#lim_btn(load_im,im_u,im)
|
165 |
+
im_btn.click(run,[im,im_u],out_im)
|
166 |
+
app.queue(concurrency_count=100).launch(show_api=False)
|