[pytorch] 2D + 3D ResNet代码实现, 改写

本文只介绍resnet的代码实现，需要对resnet有基础的了解。代码参考pytorch官方实现，删除了一些非必要的判断条件，看起来更加简洁。z再次基础上，可以根据需要加入自己需要调整的参数，比如dilation，norm_layer等.

参考
SOURCE CODE FOR TORCHVISION.MODELS.RESNET
2D ResNet50 网络结构搭建(PyTorch)
MedicalNet

网络结构

左图：BasicBlock结构，用于resnet18/34
右图： Bottleneck结构，用于resnet50/101/152

2D ResNet代码

import torch
import torch.nn as nn

首先是两种block的代码

#18/34
class BasicBlock_2d(nn.Module):
    expansion = 1 #每一个conv的卷积核个数的倍数

    def __init__(self, in_channel, out_channel, stride=1, downsample=None):#downsample对应虚线残差结构
        super(BasicBlock_2d, self).__init__()
        self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
                               kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channel)#BN处理
        self.relu = nn.ReLU(inplace=True) # 尽量使用inplace操作flag，节省显存
        self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel,
                               kernel_size=3, stride=1, padding=1, bias=False) 
        self.bn2 = nn.BatchNorm2d(out_channel)
        self.downsample = downsample

    def forward(self, x):
        identity = x #捷径上的输出值，为了保证原始输入与卷积后的输出层叠加时维度相同
        if self.downsample is not None:
            identity = self.downsample(x)

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

        out += identity
        out = self.relu(out)

        return out

#50,101,152
class Bottleneck_2d(nn.Module):
    '''

    :param in_channel: 输入block的之前的通道数
    :param out_channel: 在block中间处理的时候的通道数
            out_channel*self.extention:输出的维度
    :param stride:卷积步长
    :param downsample:在_make_layer函数中赋值，在resnet每层链接的第一个卷积层需要改变通道 如resnet50 conv2_x输出的256降低为128 conv3_x的输入
    '''
    expansion = 4 #4倍，类变量，可通过类名修改

    def __init__(self, in_channel, out_channel, stride=1, downsample=None):
        super(Bottleneck_2d, self).__init__()
        self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
                               kernel_size=1, stride=1, bias=False)  # squeeze channels
        self.bn1 = nn.BatchNorm2d(out_channel)
        # -----------------------------------------
        self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel,
                               kernel_size=3, stride=stride, bias=False, padding=1)
        self.bn2 = nn.BatchNorm2d(out_channel)
        # -----------------------------------------
        self.conv3 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel*self.expansion,#输出*4
                               kernel_size=1, stride=1, bias=False)  # unsqueeze channels
        self.bn3 = nn.BatchNorm2d(out_channel*self.expansion)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample

    def forward(self, x):
        identity = x
        if self.downsample is not None:
            identity = self.downsample(x)

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        out += identity  # 残差连接
        out = self.relu(out)

        return out

class ResNet2d(nn.Module):
    """
    __init__
        block: 堆叠的基本模块
        block_num: 基本模块堆叠个数,是一个list,对于resnet50=[3,4,6,3]
        num_classes: 全连接之后的分类特征维度
        
    _make_layer
        block: 堆叠的基本模块
        channel: 每个stage中堆叠模块的第一个卷积的卷积核个数，对resnet50分别是:64,128,256,512
        block_num: 当期stage堆叠block个数
        stride: 默认卷积步长
    """
    def __init__(self, block, blocks_num, num_classes=1000, include_top=True):#block残差结构 include_top为了之后搭建更加复杂的网络
        super(ResNet2d, self).__init__()
        self.include_top = include_top
        self.in_channel = 64   # conv1的输出维度

        self.conv1_2d = nn.Conv2d(3, self.in_channel, kernel_size=7, stride=2,
                               padding=3, bias=False)
        self.bn1_2d = nn.BatchNorm2d(self.in_channel)
        self.relu_2d = nn.ReLU(inplace=True)
        self.maxpool_2d = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) # H/2,W/2。C不变
        self.layer1_2d = self._make_layer(block, 64, blocks_num[0]) # H,W不变。downsample控制的shortcut，out_channel=64x4=256
        self.layer2_2d = self._make_layer(block, 128, blocks_num[1], stride=2) # H/2, W/2。downsample控制的shortcut，out_channel=128x4=512
        self.layer3_2d = self._make_layer(block, 256, blocks_num[2], stride=2) # H/2, W/2。downsample控制的shortcut，out_channel=256x4=1024
        self.layer4_2d = self._make_layer(block, 512, blocks_num[3], stride=2) # H/2, W/2。downsample控制的shortcut，out_channel=512x4=2048
        if self.include_top:
            self.avgpool = nn.AdaptiveAvgPool2d((1, 1))  # output size = (1, 1)自适应
            self.fc = nn.Linear(512 * block.expansion, num_classes)

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

    def _make_layer(self, block, channel, block_num, stride=1):
        """
        第一个输入参数 block 选择要使用的模块是 BasicBlock 还是 Bottleneck 类，
        第二个输入参数 channel 是该模块的输出通道数，
        第三个输入参数 block_num 是每个 blocks 中包含多少个 residual 子结构。
        """
        downsample = None  # 用于控制shorcut路的
        if stride != 1 or self.in_channel != channel * block.expansion:
        # 对resnet50：conv2中特征图尺寸H,W不需要下采样/2，但是通道数x4，
        # 因此shortcut通道数也需要x4。对其余conv3,4,5，既要特征图尺寸H,W/2，又要shortcut维度x4
            downsample = nn.Sequential(
                nn.Conv2d(self.in_channel, channel * block.expansion, kernel_size=1, stride=stride, bias=False),
                # out_channels决定输出通道数x4，stride决定特征图尺寸H,W/2
                nn.BatchNorm2d(channel * block.expansion))

        layers = []
        layers.append(block(self.in_channel, channel, downsample=downsample, stride=stride)) # 定义convi_x中的第一个残差块，只有第一个需要设置downsample和stride
        self.in_channel = channel * block.expansion  # 在下一次调用_make_layer函数的时候，self.in_channel已经x4

        for _ in range(1, block_num): # 通过循环堆叠其余残差块(堆叠了剩余的block_num-1个)
            layers.append(block(self.in_channel, channel))

        return nn.Sequential(*layers)  # '*'的作用是将list转换为非关键字参数传入

    def forward(self, x):
        x = self.conv1_2d(x)
        x = self.bn1_2d(x)
        x = self.relu_2d(x)
        x = self.maxpool_2d(x)

        x = self.layer1_2d(x)
        x = self.layer2_2d(x)
        x = self.layer3_2d(x)
        x = self.layer4_2d(x)

        if self.include_top:
            x = self.avgpool(x)
            x = torch.flatten(x, 1)
            x = self.fc(x)

        return x

使用resnet50进行3分类

resnet50_2d = ResNet2d(Bottleneck_2d, [3, 4, 6, 3], num_classes=3,include_top=True)
x=torch.randn(1,3,448,448)
X=resnet50_2d(x)
print(X.shape)
# torch.Size([1, 3])

网络结构可视化

import netron
import torch.onnx
modelData ='demo.onnx' # 定义模型数据保存的路径
torch.onnx.export(resnet50_2d, x, modelData)  # 将 pytorch 模型以 onnx 格式导出并保存
onnx.save(onnx.shape_inference.infer_shapes(onnx.load(modelData)), modelData)
netron.start(modelData)  # 输出网络结构

2D ResNet的backbones

resnet18_2d = ResNet2d(BasicBlock_2d, [2, 2, 2, 2], include_top=False)
resnet34_2d = ResNet2d(BasicBlock_2d, [3, 4, 6, 3], include_top=False)
resnet50_2d = ResNet2d(Bottleneck_2d, [3, 4, 6, 3], include_top=False)
resnet101_2d = ResNet2d(Bottleneck_2d, [3, 4, 23, 3], include_top=False)
resnet152_2d = ResNet2d(Bottleneck_2d, [3, 8, 36, 3], include_top=False)

3D ResNet代码

3D ResNet的实现参考腾讯的MedicalNet: Med3D: Transfer Learning for 3D Medical Image Analysis.

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
import math
from functools import partial

def conv3x3x3(in_planes, out_planes, stride=1, dilation=1):
    # 3x3x3 convolution with padding
    return nn.Conv3d(
        in_planes,
        out_planes,
        kernel_size=3,
        dilation=dilation,
        stride=stride,
        padding=dilation,
        bias=False)


def downsample_basic_block(x, planes, stride, no_cuda=False):
    out = F.avg_pool3d(x, kernel_size=1, stride=stride)
    zero_pads = torch.Tensor(
        out.size(0), planes - out.size(1), out.size(2), out.size(3),
        out.size(4)).zero_()
    if not no_cuda:
        if isinstance(out.data, torch.cuda.FloatTensor):
            zero_pads = zero_pads.cuda()

    out = Variable(torch.cat([out.data, zero_pads], dim=1))

    return out


class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, dilation=1, downsample=None):
        super(BasicBlock, self).__init__()
        self.conv1 = conv3x3x3(inplanes, planes, stride=stride, dilation=dilation)
        self.bn1 = nn.BatchNorm3d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3x3(planes, planes, dilation=dilation)
        self.bn2 = nn.BatchNorm3d(planes)
        self.downsample = downsample
        self.stride = stride
        self.dilation = dilation

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)
        out = self.conv2(out)
        out = self.bn2(out)

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

        out += residual
        out = self.relu(out)

        return out


class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, dilation=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv3d(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm3d(planes)
        self.conv2 = nn.Conv3d(
            planes, planes, kernel_size=3, stride=stride, dilation=dilation, padding=dilation, bias=False)
        self.bn2 = nn.BatchNorm3d(planes)
        self.conv3 = nn.Conv3d(planes, planes * 4, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm3d(planes * 4)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride
        self.dilation = dilation

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

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

        out += residual
        out = self.relu(out)

        return out

class ResNet_3d(nn.Module):

    def __init__(self,
                 block,
                 layers,
                 num_classes=1000,
                 shortcut_type='B',
                 no_cuda = False,
                 include_top=True):
        super(ResNet_3d, self).__init__()
        self.inplanes = 64
        self.no_cuda = no_cuda
        self.include_top = include_top
        
        self.conv1 = nn.Conv3d(
            1,
            64,
            kernel_size=7,
            stride=(2, 2, 2),
            padding=(3, 3, 3),
            bias=False)
            
        self.bn1 = nn.BatchNorm3d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool3d(kernel_size=(3, 3, 3), stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0], shortcut_type)
        self.layer2 = self._make_layer(
            block, 128, layers[1], shortcut_type, stride=2)
        self.layer3 = self._make_layer(
            block, 256, layers[2], shortcut_type, stride=2)
        self.layer4 = self._make_layer(
            block, 512, layers[3], shortcut_type, stride=2)
        
        if self.include_top:
            self.avgpool = nn.AdaptiveAvgPool3d((1, 1, 1))  # output size = (1, 1)自适应
            self.fc = nn.Linear(512 * block.expansion, num_classes)

        for m in self.modules():
            if isinstance(m, nn.Conv3d):
                m.weight = nn.init.kaiming_normal(m.weight, mode='fan_out')
            elif isinstance(m, nn.BatchNorm3d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()
                


    def _make_layer(self, block, planes, blocks, shortcut_type, stride=1, dilation=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            if shortcut_type == 'A':
                downsample = partial(
                    downsample_basic_block,
                    planes=planes * block.expansion,
                    stride=stride,
                    no_cuda=self.no_cuda)
            else:
                downsample = nn.Sequential(
                    nn.Conv3d(
                        self.inplanes,
                        planes * block.expansion,
                        kernel_size=1,
                        stride=stride,
                        bias=False), nn.BatchNorm3d(planes * block.expansion))

        layers = []
        layers.append(block(self.inplanes, planes, stride=stride, downsample=downsample))
        self.inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)
    

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        if self.include_top:
            x = self.avgpool(x)
            x = torch.flatten(x, 1)
            x = self.fc(x)

        return x

使用3D Resnet34进行3分类

resnet34_3d = ResNet_3d(BasicBlock, [3, 4, 6, 3],shortcut_type='A',no_cuda=False,num_classes=3,include_top=True)
x=torch.randn(1,1,256,224,164)
X=resnet34_3d(x)
print(X.shape)
# torch.Size([1, 3])

使用3D Resnet50进行3分类

resnet50_3d = ResNet_3d(Bottleneck, [3, 4, 6, 3],shortcut_type='B',no_cuda=False,num_classes=3,include_top=True)
x=torch.randn(1,1,256,224,164)
X=resnet50_3d(x)
print(X.shape)
# torch.Size([1, 3])

可视化

import netron
import torch.onnx
modelData ='best.onnx' # 定义模型数据保存的路径
torch.onnx.export(resnet50_3d, x, modelData)  # 将 pytorch 模型以 onnx 格式导出并保存
onnx.save(onnx.shape_inference.infer_shapes(onnx.load(modelData)), modelData)
netron.start(modelData)  # 输出网络结构