model.py

import torch 
import torch.nn as nn
import torchvision

import torch.nn.functional as F



device = "cuda" if torch.cuda.is_available() else "cpu"

image_transforms_rgb = torchvision.transforms.Compose([
    torchvision.transforms.Resize((256, 256)),
    torchvision.transforms.ToTensor(),
    torchvision.transforms.Normalize(mean=[0.0,0.0,0.0], std=[1.0,1.0,1.0]),
    torchvision.transforms.Grayscale()
])

image_transforms_gs = torchvision.transforms.Compose([
    torchvision.transforms.Resize((256, 256)),
    torchvision.transforms.ToTensor(),
    torchvision.transforms.Normalize(mean=[0.0], std=[1.0]),
])



class ConvBlock(nn.Module):
    def __init__(self, in_channel, out_channel):
        super(ConvBlock, self).__init__()
        self.main = nn.Sequential(
            nn.Conv2d(in_channel, out_channel, kernel_size=3, stride=1, padding=1),
            nn.BatchNorm2d(out_channel),
            nn.ReLU(True),
            nn.Conv2d(out_channel, out_channel, kernel_size=3, stride=1, padding=1),
            nn.BatchNorm2d(out_channel),
            nn.ReLU(True)
        )
        
    def forward(self, x):
        return self.main(x)

class UNETFruitColor(nn.Module):
    def __init__(self):
        super(UNETFruitColor, self).__init__()
        
        self.convs = [64, 128, 256, 512]
        self.convEncoder = nn.ModuleList()
        
        in_feature = 1
        for conv in self.convs:            
            self.convEncoder.append(ConvBlock(in_feature, conv))
            in_feature = conv
        
        self.bottleNeck = ConvBlock(self.convs[-1], self.convs[-1]*2)
        
        in_feature = self.convs[-1]*2
        
        self.convDecoder = nn.ModuleList()
        self.decoderUpConvs = nn.ModuleList()
        
        for conv in self.convs[::-1]:
            self.convDecoder.append(ConvBlock(in_feature, conv))
            self.decoderUpConvs.append(nn.ConvTranspose2d(in_feature, conv, kernel_size=2, stride=2, padding=0))
            in_feature = conv
        
        
        # final conv and deconv
        self.finalUpConv = nn.Conv2d(in_feature, 3, (1, 1))
        self.sigmoid = nn.Sigmoid()
        
    def forward(self,x):
        skip_conns = []
        for conv in self.convEncoder:
            # conv
            x = conv(x)
            # append for skip conns
            skip_conns.append(x)
            # max pool
            x = F.max_pool2d(x, (2,2), stride=2)
            
        x = self.bottleNeck(x)

        skip_conns = skip_conns[::-1]
                
        for idx in range(len(self.convDecoder)):
            # do upsample here
            upconv = self.decoderUpConvs[idx]
            deconv = self.convDecoder[idx]
            skp = skip_conns[idx]
            
            # do up conv
            x = upconv(x)
            
            # crop and cat
            x_cat = torchvision.transforms.Resize((x.shape[2], x.shape[3]))(skp)
            x = torch.cat([x_cat, x], dim=1)

            # do deconv
            x = deconv(x)
        
        # final 

        x = self.finalUpConv(x)
#         x = self.sigmoid(x)
        return x
      
      
      
model = UNETFruitColor()
model = nn.DataParallel(model).to(device)
model.load_state_dict(torch.load("unet_colorizer_flickr_5_93_Ploss_10_14K.pth", map_location=device),strict=True)
model.eval()