import sys from typing import Dict sys.path.insert(0, 'gradio-modified') import gradio as gr import numpy as np import torch.nn as nn from PIL import Image import torch if torch.cuda.is_available(): t = torch.cuda.get_device_properties(0).total_memory r = torch.cuda.memory_reserved(0) a = torch.cuda.memory_allocated(0) f = t-a # free inside reserved if f < 2**32: device = 'cpu' else: device = 'cuda' else: device = 'cpu' torch._C._jit_set_bailout_depth(0) print('Use device:', device) net = torch.jit.load(f'weights/pkp-v1.{device}.jit.pt') class BaseColor(nn.Module): def __init__(self): super(BaseColor, self).__init__() self.l_cent = 50. self.l_norm = 100. self.ab_norm = 110. def normalize_l(self, in_l): return (in_l-self.l_cent)/self.l_norm def unnormalize_l(self, in_l): return in_l*self.l_norm + self.l_cent def normalize_ab(self, in_ab): return in_ab/self.ab_norm def unnormalize_ab(self, in_ab): return in_ab*self.ab_norm class ECCVGenerator(BaseColor): def __init__(self, norm_layer=nn.BatchNorm2d): super(ECCVGenerator, self).__init__() model1=[nn.Conv2d(1, 64, kernel_size=3, stride=1, padding=1, bias=True),] model1+=[nn.ReLU(True),] model1+=[nn.Conv2d(64, 64, kernel_size=3, stride=2, padding=1, bias=True),] model1+=[nn.ReLU(True),] model1+=[norm_layer(64),] model2=[nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1, bias=True),] model2+=[nn.ReLU(True),] model2+=[nn.Conv2d(128, 128, kernel_size=3, stride=2, padding=1, bias=True),] model2+=[nn.ReLU(True),] model2+=[norm_layer(128),] model3=[nn.Conv2d(128, 256, kernel_size=3, stride=1, padding=1, bias=True),] model3+=[nn.ReLU(True),] model3+=[nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=True),] model3+=[nn.ReLU(True),] model3+=[nn.Conv2d(256, 256, kernel_size=3, stride=2, padding=1, bias=True),] model3+=[nn.ReLU(True),] model3+=[norm_layer(256),] model4=[nn.Conv2d(256, 512, kernel_size=3, stride=1, padding=1, bias=True),] model4+=[nn.ReLU(True),] model4+=[nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1, bias=True),] model4+=[nn.ReLU(True),] model4+=[nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1, bias=True),] model4+=[nn.ReLU(True),] model4+=[norm_layer(512),] model5=[nn.Conv2d(512, 512, kernel_size=3, dilation=2, stride=1, padding=2, bias=True),] model5+=[nn.ReLU(True),] model5+=[nn.Conv2d(512, 512, kernel_size=3, dilation=2, stride=1, padding=2, bias=True),] model5+=[nn.ReLU(True),] model5+=[nn.Conv2d(512, 512, kernel_size=3, dilation=2, stride=1, padding=2, bias=True),] model5+=[nn.ReLU(True),] model5+=[norm_layer(512),] model6=[nn.Conv2d(512, 512, kernel_size=3, dilation=2, stride=1, padding=2, bias=True),] model6+=[nn.ReLU(True),] model6+=[nn.Conv2d(512, 512, kernel_size=3, dilation=2, stride=1, padding=2, bias=True),] model6+=[nn.ReLU(True),] model6+=[nn.Conv2d(512, 512, kernel_size=3, dilation=2, stride=1, padding=2, bias=True),] model6+=[nn.ReLU(True),] model6+=[norm_layer(512),] model7=[nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1, bias=True),] model7+=[nn.ReLU(True),] model7+=[nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1, bias=True),] model7+=[nn.ReLU(True),] model7+=[nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1, bias=True),] model7+=[nn.ReLU(True),] model7+=[norm_layer(512),] model8=[nn.ConvTranspose2d(512, 256, kernel_size=4, stride=2, padding=1, bias=True),] model8+=[nn.ReLU(True),] model8+=[nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=True),] model8+=[nn.ReLU(True),] model8+=[nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=True),] model8+=[nn.ReLU(True),] model8+=[nn.Conv2d(256, 313, kernel_size=1, stride=1, padding=0, bias=True),] self.model1 = nn.Sequential(*model1) self.model2 = nn.Sequential(*model2) self.model3 = nn.Sequential(*model3) self.model4 = nn.Sequential(*model4) self.model5 = nn.Sequential(*model5) self.model6 = nn.Sequential(*model6) self.model7 = nn.Sequential(*model7) self.model8 = nn.Sequential(*model8) self.softmax = nn.Softmax(dim=1) self.model_out = nn.Conv2d(313, 2, kernel_size=1, padding=0, dilation=1, stride=1, bias=False) self.upsample4 = nn.Upsample(scale_factor=4, mode='bilinear') def forward(self, input_l): conv1_2 = self.model1(self.normalize_l(input_l)) conv2_2 = self.model2(conv1_2) conv3_3 = self.model3(conv2_2) conv4_3 = self.model4(conv3_3) conv5_3 = self.model5(conv4_3) conv6_3 = self.model6(conv5_3) conv7_3 = self.model7(conv6_3) conv8_3 = self.model8(conv7_3) out_reg = self.model_out(self.softmax(conv8_3)) return self.unnormalize_ab(self.upsample4(out_reg)) # model_net = torch.load(f'weights/colorizer.pt') model_net = ECCVGenerator() model_net.load_state_dict(torch.load(f'weights/colorizer (1).pt', map_location=torch.device('cpu'))) def resize_original(img: Image.Image): if img is None: return img if isinstance(img, dict): img = img["image"] guide_img = img.convert('L') w, h = guide_img.size scale = 256 / min(guide_img.size) guide_img = guide_img.resize([int(round(s*scale)) for s in guide_img.size], Image.Resampling.LANCZOS) guide = np.asarray(guide_img) h, w = guide.shape[-2:] rows = int(np.ceil(h/64))*64 cols = int(np.ceil(w/64))*64 ph_1 = (rows-h) // 2 ph_2 = rows-h - (rows-h) // 2 pw_1 = (cols-w) // 2 pw_2 = cols-w - (cols-w) // 2 guide = np.pad(guide, ((ph_1, ph_2), (pw_1, pw_2)), mode='constant', constant_values=255) guide_img = Image.fromarray(guide) return gr.Image.update(value=guide_img.convert('RGBA')), guide_img.convert('RGBA') def resize_original2(img: Image.Image): if img is None: return img if isinstance(img, dict): img = img["image"] img = img.resize(256,256) return img def colorize(img: Dict[str, Image.Image], guide_img: Image.Image, seed: int, hint_mode: str): if not isinstance(img, dict): return gr.update(visible=True) if hint_mode == "Roughly Hint": hint_mode_int = 0 elif hint_mode == "Precisely Hint": hint_mode_int = 0 guide_img = guide_img.convert('L') hint_img = img["mask"].convert('RGBA') # I modified gradio to enable it upload colorful mask guide = torch.from_numpy(np.asarray(guide_img))[None,None].float().to(device) / 255.0 * 2 - 1 hint = torch.from_numpy(np.asarray(hint_img)).permute(2,0,1)[None].float().to(device) / 255.0 * 2 - 1 hint_alpha = (hint[:,-1:] > 0.99).float() hint = hint[:,:3] * hint_alpha - 2 * (1 - hint_alpha) np.random.seed(int(seed)) b, c, h, w = hint.shape h //= 8 w //= 8 noises = [torch.from_numpy(np.random.randn(b, c, h, w)).float().to(device) for _ in range(16+1)] with torch.inference_mode(): sample = net(noises, guide, hint, hint_mode_int) out = sample[0].cpu().numpy().transpose([1,2,0]) out = np.uint8(((out + 1) / 2 * 255).clip(0,255)) return Image.fromarray(out).convert('RGB') def colorize2(img: Image.Image, model_option: str): if not isinstance(img, dict): return gr.update(visible=True) if hint_mode == "Model 1": model_int = 0 elif hint_mode == "Model 2": model_int = 0 with torch.inference_mode(): out2 = model_net(input) out2 = sample[0].cpu().numpy().transpose([1,2,0]) out2 = np.uint8(((out + 1) / 2 * 255).clip(0,255)) return Image.fromarray(out2).convert('RGB') with gr.Blocks() as demo: gr.Markdown('''