关于理解Pytorch.named_modules()循环的问题

Question

I am referring to this implementation here:

https://github.com/hszhao/semseg/blob/master/model/pspnet.py

In line 49-58, the author writes:

for n, m in self.layer3.named_modules():
    if 'conv2' in n:
        m.dilation, m.padding, m.stride = (2, 2), (2, 2), (1, 1)
    elif 'downsample.0' in n:
        m.stride = (1, 1)

for n, m in self.layer4.named_modules():
    if 'conv2' in n:
        m.dilation, m.padding, m.stride = (4, 4), (4, 4), (1, 1)
    elif 'downsample.0' in n:
        m.stride = (1, 1)

What exactly is happening in these loops?

My understanding is, that the author is creatig a resnet model (his resnet.py here https://github.com/hszhao/semseg/blob/master/model/resnet.py ) and then is calling the different layers, which he implemented in his resnet class to forward them below.

layer3 and layer4 in resnet.py are made by calling the function def _make_layer(self, block, planes, blocks, stride=1): , so I assume that when .named_modules() is used in the loops, it is looping through the modules in this def _make_layer function, is it? If so, what happens in the elif part? There is no module, that is called downsample.0 ? (The only modules are nn.Conv2d and nn.BatchNorm2d )

Answer 1

Below is an example of resnet that used there.

model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)

In Resnet class, it calls super because of this, it has self.downsample If it's not none:

if self.downsample is not None:
        residual = self.downsample(x)

it could have Sequential or another layer.

class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
                               padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)
        self.conv3 = nn.Conv2d(planes, planes * self.expansion, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes * self.expansion)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride