Question about resnet.py dimension

keunwoo · July 3, 2018, 5:56am

I’m new to PyTorch and deep learning.
While reading resnet.py code, there was a line that I couldn’t understand.

pytorch/vision/blob/1fb0ccf71620d113cb72696b2eb8317b3e252cbb/torchvision/models/resnet.py#L111


    self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
                           bias=False)
    self.bn1 = nn.BatchNorm2d(64)
    self.relu = nn.ReLU(inplace=True)
    self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
    self.layer1 = self._make_layer(block, 64, layers[0])
    self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
    self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
    self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
    self.avgpool = nn.AvgPool2d(7, stride=1)
    self.fc = nn.Linear(512 * block.expansion, num_classes)


    for m in self.modules():
        if isinstance(m, nn.Conv2d):
            nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
        elif isinstance(m, nn.BatchNorm2d):
            nn.init.constant_(m.weight, 1)
            nn.init.constant_(m.bias, 0)


def _make_layer(self, block, planes, blocks, stride=1):
    downsample = None

Why is fc layer not considering the size of an image?
Shouldn’t it consider not only feature size, but also width and height of output of the avgpool layer?

Naman-ntc · July 3, 2018, 5:59am

The avg pooling layer just before the fully connected reduces it to 1x1 image size so you dont need that

Naman-ntc · July 3, 2018, 7:47am

It doesn’t ensure this, you’ll have to manuualy change the fully connected if you choose other input size than 224x224