Hello, I try to build ResNet18 from scratch (similar to the archtecture of the library), though for some reason it does not behave as I want to.
In the first residual block (ResidualBlock), I dont want the downsampling layers to exist however, when I print the model, they are shown in the architecture.
Am I missing something?
Here is the residual block:
class ResidualBlock(nn.Module):
def init(self, in_channels, out_channels, stride = 1, downsample = False):
super(ResidualBlock, self).init()self.conv = nn.Sequential( nn.Conv2d(in_channels, out_channels, kernel_size = 3, stride = stride, padding = 1), nn.BatchNorm2d(out_channels), nn.ReLU(), nn.Conv2d(in_channels, out_channels, kernel_size = 3, stride = stride, padding = 1), nn.BatchNorm2d(out_channels)) self.downsample = downsample print('--------------', self.downsample) self.downsample_layers = nn.Sequential( nn.Conv2d(in_channels, out_channels, kernel_size = 1, stride = 2), nn.BatchNorm2d(out_channels)) self.relu = nn.ReLU() self.out_channels = out_channels def forward(self, x): residual = x out = self.conv(x) if self.downsample: residual = self.downsample_layers(x) out += residual out = self.relu(out) return out
Here is the ResNet:
class ResNet(nn.Module):
def init(self, block, layers, num_classes = 10):
super(ResNet, self).init()self.inplanes = 64 self.conv1 = nn.Sequential( nn.Conv2d(3, 64, kernel_size = 7, stride = 2, padding = 3), nn.BatchNorm2d(64), nn.ReLU(), nn.MaxPool2d(kernel_size = 3, stride = 2, padding = 1)) self.layer0 = self._make_layer(block, 64, layers[0], stride = 1, downsample = False) self.layer1 = self._make_layer(block, 128, layers[1], stride = 2, downsample = True) self.layer2 = self._make_layer(block, 256, layers[2], stride = 2, downsample = True) self.layer3 = self._make_layer(block, 512, layers[3], stride = 2, downsample = True) self.avgpool = nn.AvgPool2d(7, stride=1) self.fc = nn.Linear(512, num_classes) def _make_layer(self, block, planes, blocks, stride=1, downsample = False): layers = [] if downsample: layers.append(block(self.inplanes, planes, stride, downsample=True)) else: layers.append(block(self.inplanes, planes, stride, downsample=False)) self.inplanes = planes return nn.Sequential(*layers) def forward(self, 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.fc(x) return x
I have based my code on this article: Writing ResNet from Scratch in PyTorch which works but I wanted to changed it a bit.
Thank you