Update pintar.py
Browse files
pintar.py
CHANGED
@@ -20,7 +20,7 @@ def Lab2RGB_out(img_lab):
|
|
20 |
img_ab = img_lab[:,1:,:,:]
|
21 |
img_l = img_l + 50
|
22 |
pred_lab = torch.cat((img_l, img_ab), 1)[0,...].numpy()
|
23 |
-
out = (np.clip(color.lab2rgb(pred_lab.transpose(1, 2, 0)), 0, 1)
|
24 |
return out
|
25 |
|
26 |
def RGB2Lab(inputs):
|
@@ -34,11 +34,11 @@ def Normalize(inputs):
|
|
34 |
return lab.astype('float32')
|
35 |
|
36 |
def numpy2tensor(inputs):
|
37 |
-
out = torch.from_numpy(inputs.transpose(2,
|
38 |
return out
|
39 |
|
40 |
def tensor2numpy(inputs):
|
41 |
-
out = inputs[0
|
42 |
return out
|
43 |
|
44 |
def preprocessing(inputs):
|
@@ -49,41 +49,139 @@ def preprocessing(inputs):
|
|
49 |
return img.unsqueeze(0), img_lab.unsqueeze(0)
|
50 |
|
51 |
if __name__ == "__main__":
|
52 |
-
parser = argparse.ArgumentParser()
|
53 |
-
parser.add_argument("-r", "--reference", type=str, help="ruta de la imagen de referencia")
|
54 |
-
parser.add_argument("-o", "--output", type=str, help="carpeta de salida para las im谩genes coloreadas")
|
55 |
-
parser.add_argument("-ckpt", "--model_checkpoint", type=str, help="ruta del modelo de checkpoint")
|
56 |
-
args = parser.parse_args()
|
57 |
-
|
58 |
device = "cuda"
|
59 |
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
while True:
|
67 |
-
# ... (resto del c贸digo)
|
68 |
-
|
69 |
-
with torch.no_grad():
|
70 |
-
img2_resize = F.interpolate(img2 / 255., size=(256, 256), mode='bilinear', recompute_scale_factor=False, align_corners=False)
|
71 |
-
img1_L_resize = F.interpolate(img1_lab[:,:1,:,:] / 50., size=(256, 256), mode='bilinear', recompute_scale_factor=False, align_corners=False)
|
72 |
-
|
73 |
-
color_vector = colorEncoder(img2_resize)
|
74 |
-
|
75 |
-
fake_ab = colorUNet((img1_L_resize, color_vector))
|
76 |
-
fake_ab = F.interpolate(fake_ab * 110, size=(height, width), mode='bilinear', recompute_scale_factor=False, align_corners=False)
|
77 |
-
|
78 |
-
fake_img = torch.cat((img1_lab[:,:1,:,:], fake_ab), 1)
|
79 |
-
fake_img = Lab2RGB_out(fake_img)
|
80 |
|
81 |
-
|
82 |
-
if not os.path.exists(out_folder):
|
83 |
-
os.makedirs(out_folder)
|
84 |
-
out_img_path = os.path.join(out_folder, f'{img_name}_color.png')
|
85 |
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
img_ab = img_lab[:,1:,:,:]
|
21 |
img_l = img_l + 50
|
22 |
pred_lab = torch.cat((img_l, img_ab), 1)[0,...].numpy()
|
23 |
+
out = (np.clip(color.lab2rgb(pred_lab.transpose(1, 2, 0)), 0, 1)* 255).astype("uint8")
|
24 |
return out
|
25 |
|
26 |
def RGB2Lab(inputs):
|
|
|
34 |
return lab.astype('float32')
|
35 |
|
36 |
def numpy2tensor(inputs):
|
37 |
+
out = torch.from_numpy(inputs.transpose(2,0,1))
|
38 |
return out
|
39 |
|
40 |
def tensor2numpy(inputs):
|
41 |
+
out = inputs[0,...].detach().cpu().numpy().transpose(1,2,0)
|
42 |
return out
|
43 |
|
44 |
def preprocessing(inputs):
|
|
|
49 |
return img.unsqueeze(0), img_lab.unsqueeze(0)
|
50 |
|
51 |
if __name__ == "__main__":
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
device = "cuda"
|
53 |
|
54 |
+
parser = argparse.ArgumentParser()
|
55 |
+
parser.add_argument("--path", type=str, default=None, help="path of input image")
|
56 |
+
parser.add_argument("--size", type=int, default=None)
|
57 |
+
parser.add_argument("--ckpt", type=str, default=None, help="path of model weight")
|
58 |
+
parser.add_argument("-ne", "--no_extractor", action='store_true', help="Do not segment the manga panels.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
|
60 |
+
args = parser.parse_args()
|
|
|
|
|
|
|
61 |
|
62 |
+
if args.path:
|
63 |
+
test_dir_path = args.path
|
64 |
+
if args.size:
|
65 |
+
imgsize = args.size
|
66 |
+
if args.ckpt:
|
67 |
+
ckpt_path = args.ckpt
|
68 |
+
if args.no_extractor:
|
69 |
+
no_extractor = args.no_extractor
|
70 |
+
|
71 |
+
ckpt = torch.load(ckpt_path, map_location=lambda storage, loc: storage)
|
72 |
+
|
73 |
+
colorEncoder = ColorEncoder().to(device)
|
74 |
+
colorEncoder.load_state_dict(ckpt["colorEncoder"])
|
75 |
+
colorEncoder.eval()
|
76 |
+
|
77 |
+
colorUNet = ColorUNet().to(device)
|
78 |
+
colorUNet.load_state_dict(ckpt["colorUNet"])
|
79 |
+
colorUNet.eval()
|
80 |
+
|
81 |
+
imgs = []
|
82 |
+
imgs_lab = []
|
83 |
+
|
84 |
+
while 1:
|
85 |
+
print(f'make sure both manga image and reference images are under this path {test_dir_path}')
|
86 |
+
img_path = input("please input the name of image needed to be colorized (with file extension): ")
|
87 |
+
img_path = os.path.join(test_dir_path, img_path)
|
88 |
+
img_name = os.path.basename(img_path)
|
89 |
+
img_name = os.path.splitext(img_name)[0]
|
90 |
+
|
91 |
+
if no_extractor:
|
92 |
+
ref_img_path = os.path.join(test_dir_path, input(f"Enter the reference image path: "))
|
93 |
+
|
94 |
+
img1 = Image.open(img_path).convert("RGB")
|
95 |
+
width, height = img1.size
|
96 |
+
img2 = Image.open(ref_img_path).convert("RGB")
|
97 |
+
|
98 |
+
img1, img1_lab = preprocessing(img1)
|
99 |
+
img2, img2_lab = preprocessing(img2)
|
100 |
+
|
101 |
+
img1 = img1.to(device)
|
102 |
+
img1_lab = img1_lab.to(device)
|
103 |
+
img2 = img2.to(device)
|
104 |
+
img2_lab = img2_lab.to(device)
|
105 |
+
|
106 |
+
with torch.no_grad():
|
107 |
+
img2_resize = F.interpolate(img2 / 255., size=(imgsize, imgsize), mode='bilinear',
|
108 |
+
recompute_scale_factor=False, align_corners=False)
|
109 |
+
img1_L_resize = F.interpolate(img1_lab[:, :1, :, :] / 50., size=(imgsize, imgsize), mode='bilinear',
|
110 |
+
recompute_scale_factor=False, align_corners=False)
|
111 |
+
|
112 |
+
color_vector = colorEncoder(img2_resize)
|
113 |
+
|
114 |
+
fake_ab = colorUNet((img1_L_resize, color_vector))
|
115 |
+
fake_ab = F.interpolate(fake_ab * 110, size=(height, width), mode='bilinear',
|
116 |
+
recompute_scale_factor=False, align_corners=False)
|
117 |
+
|
118 |
+
fake_img = torch.cat((img1_lab[:, :1, :, :], fake_ab), 1)
|
119 |
+
fake_img = Lab2RGB_out(fake_img)
|
120 |
+
|
121 |
+
out_folder = os.path.dirname(img_path)
|
122 |
+
out_name = os.path.basename(img_path)
|
123 |
+
out_name = os.path.splitext(out_name)[0]
|
124 |
+
out_img_path = os.path.join(out_folder, 'color', f'{out_name}_color.png')
|
125 |
+
|
126 |
+
# show image
|
127 |
+
Image.fromarray(fake_img).show()
|
128 |
+
# save image
|
129 |
+
folder_path = os.path.join(out_folder, 'color')
|
130 |
+
if not os.path.exists(folder_path):
|
131 |
+
os.makedirs(folder_path)
|
132 |
+
io.imsave(out_img_path, fake_img)
|
133 |
+
|
134 |
+
continue
|
135 |
+
|
136 |
+
panel_extractor = PanelExtractor(min_pct_panel=5, max_pct_panel=90)
|
137 |
+
panels, masks, panel_masks = panel_extractor.extract(img_path)
|
138 |
+
panel_num = len(panels)
|
139 |
+
|
140 |
+
ref_img_paths = []
|
141 |
+
print("Please enter the name of the reference image in order according to the number prompts on the picture")
|
142 |
+
for i in range(panel_num):
|
143 |
+
ref_img_path = os.path.join(test_dir_path, input(f"{i+1}/{panel_num} reference image:"))
|
144 |
+
ref_img_paths.append(ref_img_path)
|
145 |
+
|
146 |
+
fake_imgs = []
|
147 |
+
for i in range(panel_num):
|
148 |
+
img1 = Image.fromarray(panels[i]).convert("RGB")
|
149 |
+
width, height = img1.size
|
150 |
+
img2 = Image.open(ref_img_paths[i]).convert("RGB")
|
151 |
+
|
152 |
+
img1, img1_lab = preprocessing(img1)
|
153 |
+
img2, img2_lab = preprocessing(img2)
|
154 |
+
|
155 |
+
img1 = img1.to(device)
|
156 |
+
img1_lab = img1_lab.to(device)
|
157 |
+
img2 = img2.to(device)
|
158 |
+
img2_lab = img2_lab.to(device)
|
159 |
+
|
160 |
+
with torch.no_grad():
|
161 |
+
img2_resize = F.interpolate(img2 / 255., size=(imgsize, imgsize), mode='bilinear', recompute_scale_factor=False, align_corners=False)
|
162 |
+
img1_L_resize = F.interpolate(img1_lab[:,:1,:,:] / 50., size=(imgsize, imgsize), mode='bilinear', recompute_scale_factor=False, align_corners=False)
|
163 |
+
|
164 |
+
color_vector = colorEncoder(img2_resize)
|
165 |
+
|
166 |
+
fake_ab = colorUNet((img1_L_resize, color_vector))
|
167 |
+
fake_ab = F.interpolate(fake_ab*110, size=(height, width), mode='bilinear', recompute_scale_factor=False, align_corners=False)
|
168 |
+
|
169 |
+
fake_img = torch.cat((img1_lab[:,:1,:,:], fake_ab), 1)
|
170 |
+
fake_img = Lab2RGB_out(fake_img)
|
171 |
+
fake_imgs.append(fake_img)
|
172 |
+
|
173 |
+
if panel_num == 1:
|
174 |
+
out_folder = os.path.dirname(img_path)
|
175 |
+
out_name = os.path.basename(img_path)
|
176 |
+
out_name = os.path.splitext(out_name)[0]
|
177 |
+
out_img_path = os.path.join(out_folder,'color',f'{out_name}_color.png')
|
178 |
+
|
179 |
+
Image.fromarray(fake_imgs[0]).show()
|
180 |
+
folder_path = os.path.join(out_folder, 'color')
|
181 |
+
if not os.path.exists(folder_path):
|
182 |
+
os.makedirs(folder_path)
|
183 |
+
io.imsave(out_img_path, fake_imgs[0])
|
184 |
+
else:
|
185 |
+
panel_extractor.concatPanels(img_path, fake_imgs, masks, panel_masks)
|
186 |
+
|
187 |
+
print(f'Colored images have been saved to: {os.path.join(test_dir_path, "color")}')
|