resnet18 paddle2.2实现

resnet的特点在于残差结构，可以有效的防止梯度消失。对x+f(x)求导，可以得到1+f’(x)，所以梯度是常数。从工程实现上，可以得到的启示包括，目前的网络设计的基本卷积结构都是由一个卷积层+bn层+激活函数 构成的小模块。其他参数都是调优的结果。本文resnet18的实现是基于paddle2.2的版本

import paddle
import paddle.nn as nn

#paddle.set_device('cpu')
class Identity(nn.Layer):
    def __init_(self):
        super().__init__()

    def forward(self, x):
        return x


class Block(nn.Layer):
    def __init__(self, in_dim, out_dim, stride):
        super().__init__()
        ## 补充代码
        self.conv1 = nn.Conv2D(in_dim, out_dim, 3, stride, 1)
        self.bn1 = nn.BatchNorm2D(out_dim)
        self.conv2 = nn.Conv2D(out_dim, out_dim, 3, 1, 1)
        self.bn2 = nn.BatchNorm2D(out_dim)
        self.relu = nn.ReLU()
        if stride != 1:
            downsample = []
            downsample.append(nn.Conv2D(in_dim, out_dim, 3, stride, 1))
            downsample.append(nn.BatchNorm2D(out_dim))
            self.downsample = nn.Sequential(*downsample)
        else:
            self.downsample = Identity()
    def forward(self, x):
        ## 补充代码
        h = x
        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)
        out = self.conv2(out)
        out = self.bn2(out)
        identity = self.downsample(h)
        out += identity
        out = self.relu(out)
        return out

class ResNet18(nn.Layer):
    def __init__(self, in_dim=64, num_classes=1000):
        super().__init__()
        ## 补充代码
        self.in_dim = in_dim
        self.conv = nn.Conv2D(3, 64, 7, 2)
        self.norm = nn.BatchNorm2D(64)
        self.relu = nn.ReLU()
        self.max_pooling = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(64, 2, 1)
        self.layer2 = self._make_layer(128, 2, 2)
        self.layer3 = self._make_layer(256, 2, 2)
        self.layer4 = self._make_layer(512, 2, 2)
        self.avgpool = nn.AdaptiveAvgPool2D((1, 1))
        self.fc = nn.Linear(512 , num_classes)


    def _make_layer(self, out_dim, n_blocks, stride):
        ## 补充代码
        layers = []
        layers.append(Block(self.in_dim, out_dim, 2))
        self.in_dim = out_dim
        for _ in range(1, 2):
            layers.append(Block(self.in_dim, out_dim, 1))
        return nn.Sequential(*layers)

    def forward(self, x):
        ## 补充代码
        x = self.conv(x)
        x = self.norm(x)
        x = self.relu(x)
        x = self.max_pooling(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.avgpool(x)
        x = paddle.flatten(x, 1)
        x = self.fc(x)

        return x


def main():
    model = ResNet18()
    print(model)
    x = paddle.randn([2, 3, 32, 32])
    out = model(x)
    print(out.shape)

if __name__ == "__main__":
    main()