models/resnet_model_mask.py

import torch
import torch.nn as  nn
import torch.nn.functional as F

class ResidualBlock(nn.Module):
    def __init__(self, in_channels, out_channels, stride = 1, downsample = None):
        super(ResidualBlock, self).__init__()
        self.conv1 = nn.Sequential(
                        nn.Conv2d(in_channels, out_channels, kernel_size = 3, stride = stride, padding = 1),
                        nn.BatchNorm2d(out_channels),
                        nn.ReLU())
        self.conv2 = nn.Sequential(
                        nn.Conv2d(out_channels, out_channels, kernel_size = 3, stride = 1, padding = 1),
                        nn.BatchNorm2d(out_channels))
        self.downsample = downsample
        self.relu = nn.ReLU()
        self.out_channels = out_channels
        self.dropout_percentage = 0.5
        self.dropout1 = nn.Dropout(p=self.dropout_percentage)
        self.batchnorm_mod = nn.BatchNorm2d(out_channels)
        
    def forward(self, x):
        residual = x
        out = self.conv1(x)
        out = self.dropout1(out)
        # out = self.batchnorm_mod(out)
        out = self.conv2(out)
        out = self.dropout1(out)
        # out = self.batchnorm_mod(out)
        if self.downsample:
            residual = self.downsample(x)
        out += residual
        out = self.relu(out)
        return out


class ResNet(nn.Module):
    def __init__(self, inchan, block, layers, num_classes = 10):
        super(ResNet, self).__init__()
        self.inplanes = 64
        self.eps = 1e-5
        self.relu = nn.ReLU()
        self.conv1 = nn.Sequential(
                        nn.Conv2d(inchan, 64, kernel_size = 7, stride = 2, padding = 3),
                        nn.BatchNorm2d(64),
                        nn.ReLU())
        self.maxpool = nn.MaxPool2d(kernel_size = (2, 2), stride = 2, padding = 1)
        self.layer0 = self._make_layer(block, 64, layers[0], stride = 1)
        self.layer1 = self._make_layer(block, 128, layers[1], stride = 2)
        self.layer2 = self._make_layer(block, 256, layers[2], stride = 2)
        self.layer3 = self._make_layer(block, 512, layers[3], stride = 1)
        self.avgpool = nn.AvgPool2d(7, stride=1)
        self.fc = nn.Linear(39424, num_classes)
        self.dropout_percentage = 0.3
        self.dropout1 = nn.Dropout(p=self.dropout_percentage)

        # Encoder
        self.encoder = nn.Sequential(
            nn.Conv2d(24, 32,  kernel_size = 3, stride =1, padding = 1),
            nn.ReLU(True),nn.Dropout(p=self.dropout_percentage),
            nn.Conv2d(32, 64,  kernel_size = 3, stride =1, padding = 1),
            nn.ReLU(True),nn.Dropout(p=self.dropout_percentage),
            nn.Conv2d(64, 32,  kernel_size = 3, stride = 1, padding = 1),
            nn.ReLU(True),nn.Dropout(p=self.dropout_percentage),
            nn.Conv2d(32, 24, kernel_size = 3, stride = 1, padding = 1),
            nn.Sigmoid()
        )
        params = sum(p.numel() for p in self.encoder.parameters())
        print("num params encoder ",params)

    def norm(self, x):
        shifted = x-x.min()
        maxes = torch.amax(abs(shifted), dim=(-2, -1))
        repeated_maxes = maxes.unsqueeze(2).unsqueeze(3).repeat(1, 1, x.shape[-2],x.shape[-1])
        x = shifted/repeated_maxes
        return x

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes, kernel_size=1, stride=stride),
                nn.BatchNorm2d(planes),
            )
        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes))
        return nn.Sequential(*layers)
    
    def forward(self, x, return_mask=False):
        # # m = self.encoder(x).unsqueeze(-1).repeat(1, 1, 1, x.shape[-1])
        m = self.encoder(x)
        self.mask = m
        self.value = x
        # # m = nn.Sigmoid()(self.encoder(x))
        x = x * m
        # x = self.norm(x)
        x = self.conv1(x)
        x = self.maxpool(x)
        x = self.layer0(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.dropout1(x)
        x = self.fc(x)
        return x
        # if return_mask:
        #     return x, self.mask, self.value
        # else:
        #     return x


class ConvAutoencoder(nn.Module):
    def __init__(self):
        super(ConvAutoencoder, self).__init__()
        
        # Encoder
        self.encoder = nn.Sequential(
            nn.Conv2d(3, 16, kernel_size=3, stride=2, padding=1),  # (16, 96, 128)
            nn.ReLU(),
            nn.Conv2d(16, 32, kernel_size=3, stride=2, padding=1), # (32, 48, 64)
            nn.ReLU(),
            nn.Conv2d(32, 64, kernel_size=3, stride=2, padding=1), # (64, 24, 32)
            nn.ReLU(),
            nn.Conv2d(64, 128, kernel_size=3, stride=2, padding=1),# (128, 12, 16)
            nn.ReLU()
        )
        
        # Fully connected latent space
        self.fc1 = nn.Linear(128 * 12 * 16, 8)
        self.fc2 = nn.Linear(8, 128 * 12 * 16)
        
        # Decoder
        self.decoder = nn.Sequential(
            nn.ConvTranspose2d(128, 64, kernel_size=3, stride=2, padding=1, output_padding=1), # (64, 24, 32)
            nn.ReLU(),
            nn.ConvTranspose2d(64, 32, kernel_size=3, stride=2, padding=1, output_padding=1),  # (32, 48, 64)
            nn.ReLU(),
            nn.ConvTranspose2d(32, 16, kernel_size=3, stride=2, padding=1, output_padding=1),  # (16, 96, 128)
            nn.ReLU(),
            nn.ConvTranspose2d(16, 3, kernel_size=3, stride=2, padding=1, output_padding=1),   # (3, 192, 256)
            nn.Sigmoid()  # Using Sigmoid for the final activation to get output in range [0, 1]
        )
        
    def forward(self, x):
        # Encode
        x = self.encoder(x)
        
        # Flatten the encoded output
        x = x.view(x.size(0), -1)
        
        # Fully connected latent space
        x = self.fc1(x)
        x = self.fc2(x)
        
        # Reshape the output to the shape suitable for the decoder
        x = x.view(x.size(0), 128, 12, 16)
        
        # Decode
        x = self.decoder(x)
        
        return x