[人工智能]Person_reID_baseline_pytorch 源码解析之 model.py

一、基础 Resnet50 模型

1. Resnet50 模型结构

论文 Deep Residual Learning for Image Recognition 中提出了 Resnet50 分类模型。Resnet50 最初是在有 1000 个类别的 ImageNet 数据集上进行训练的，它的具体结构如下：

2. pytorch 实现的 Resnet50

导入 pytorch 实现的 resnet50，并查看网络结构

from torchvision import models
model = models.resnet50(pretrained=True)
print(model)

Resnet 50 的网络结构如下：

ResNet(
  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (relu): ReLU(inplace=True)
  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
  (layer1): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer2): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (3): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer3): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (3): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (4): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (5): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer4): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
  (fc): Linear(in_features=2048, out_features=1000, bias=True)
)

二、Reid 模型

1. 基于 Resnet50 的 Reid 模型

行人重识别问题可以看作分类问题，每一个行人 ID 作为一个类别。那么有多少个行人 ID 就有多少个行人类别。
我们在包含 751 个行人的 Reid 数据集 Market1501 上训练模型，因此需要对模型进行针对性的调整。将模型的输出类别由 1000 类调整为 751 类。
从头训练模型收敛比较困难，因此这里加载 pytorch 中预训练好的 Resnet50 模型。

import torch
import torch.nn as nn
from torchvision import models

# Define the ResNet50-based Model
class ft_net(nn.Module):
    def __init__(self, class_num = 751):
        super(ft_net, self).__init__()
        #load the model
        model_ft = models.resnet50(pretrained=True) 
        # change avg pooling to global pooling
        model_ft.avgpool = nn.AdaptiveAvgPool2d((1,1))
        self.model = model_ft
        self.classifier = ClassBlock(2048, class_num) #define our classifier.

    def forward(self, x):
        x = self.model.conv1(x)
        x = self.model.bn1(x)
        x = self.model.relu(x)
        x = self.model.maxpool(x)
        x = self.model.layer1(x)
        x = self.model.layer2(x)
        x = self.model.layer3(x)
        x = self.model.layer4(x)
        x = self.model.avgpool(x)
        x = torch.squeeze(x)
        x = self.classifier(x) #use our classifier.
        return x

然后加入一个 ClassBlock 对模型的类别数进行调整。ClassBlock 中定义了新的全连接层 fc 和 分类层 classification layer 。
fc 层|--Linear--|--bn--|--relu--| 的输入维度和输出维度在 nn.Linear(input_dim, num_bottleneck) 中定义。
classification 层的输入维度和输出维度在 nn.Linear(num_bottleneck, class_num) 中定义。

# Defines the new fc layer and classification layer
# |--Linear--|--bn--|--relu--|--Linear--|
class ClassBlock(nn.Module):
    def __init__(self, input_dim, class_num, droprate, relu=False, bnorm=True, num_bottleneck=512, linear=True, return_f = False):
        super(ClassBlock, self).__init__()
        self.return_f = return_f
        add_block = []
        if linear:
            num_bottleneck = linear
            add_block += [nn.Linear(input_dim, num_bottleneck)]
        else:
            num_bottleneck = input_dim
        if bnorm:
            add_block += [nn.BatchNorm1d(num_bottleneck)]
        if relu:
            add_block += [nn.LeakyReLU(0.1)]
        if droprate>0:
            add_block += [nn.Dropout(p=droprate)]
        add_block = nn.Sequential(*add_block)
        add_block.apply(weights_init_kaiming)

        classifier = []
        classifier += [nn.Linear(num_bottleneck, class_num)]
        classifier = nn.Sequential(*classifier)
        classifier.apply(weights_init_classifier)

        self.add_block = add_block
        self.classifier = classifier
    def forward(self, x):
        x = self.add_block(x)
        if self.return_f:
            f = x
            x = self.classifier(x)
            return [x,f]
        else:
            x = self.classifier(x)
            return x

其中，add_block.apply(weights_init_kaiming) 是指在 add_block 所有子模块上递归地应用 weights_init_kaiming 函数
classifier.apply(weights_init_classifier) 是指在 classifier 所有子模块上递归地应用 weights_init_classifier 函数，
从而使用 torch.nn.Module.apply(fn) 实现模型权重的初始化。

2. 模型权重初始化

如果不指定权重初始化方式，模型将被随机初始化。那么为什么要进行权重初始化呢？

权重初始化的目的是防止在深度神经网络的正向（前向）传播过程中层激活函数的输出损失梯度出现爆炸或消失。如果发生任何一种情况，损失梯度太大或太小，就无法有效地向后传播，并且即便可以向后传播，网络也需要花更长时间来达到收敛。

为了梯度有效传播和模型快速收敛，对基于 Resnet50 的 Reid 模型进行权重初始化。模型主要包含两部分：model_ft 和 classifier 。 model_ft 使用 Resnet50 在 ImageNet 上训练得到的对应权重进行初始化。 classifier 使用 add_block.apply(weights_init_kaiming) 和 classifier.apply(weights_init_classifier) 实现权重初始化。

import torch
import torch.nn as nn
from torch.nn import init

def weights_init_kaiming(m):
    classname = m.__class__.__name__
    # print(classname)
    if classname.find('Conv') != -1:
        init.kaiming_normal_(m.weight.data, a=0, mode='fan_in') # For old pytorch, you may use kaiming_normal.
    elif classname.find('Linear') != -1:
        init.kaiming_normal_(m.weight.data, a=0, mode='fan_out')
    elif classname.find('BatchNorm1d') != -1:
        init.normal_(m.weight.data, 1.0, 0.02)
    if hasattr(m, 'bias') and m.bias is not None:
        init.constant_(m.bias.data, 0.0)

def weights_init_classifier(m):
    classname = m.__class__.__name__
    if classname.find('Linear') != -1:
        init.normal_(m.weight.data, std=0.001)
        init.constant_(m.bias.data, 0.0)

3. pytorch 权重初始化相关函数

1. kaiming 正态分布初始化
   根据论文 Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification- He, K. et al. (2015) 中提出的方法，生成满足正态分布的随机值填充输入张量，并返回服从 N(0,std2) 分布的张量

其中，

torch.nn.init.kaiming_normal_(tensor, a=0, mode='fan_in', nonlinearity='leaky_relu')

• tensor 是输入张量
• mode 可以为’fan_in’（默认）或’fan_out_’。当 mode 为 ‘fan_in’ 时，保存前向传播过程中的权值变化大小，当 mode 为 ‘fan_out’ 时保存反向传播过程中的权值变化大小
• a 为当前层后使用的非线性层的负斜率，默认为 0
• nonlinearity 为非线性函数

2. 常量初始化
   用给定常量值填充输入张量，并返回张量

torch.nn.init.constant_(tensor, val)

• tensor 是输入张量
• val 是给定值

3. 正态分布初始化
   生成满足正态分布的随机值填充输入张量，并返回服从 N(mean, std2) 分布的张量

torch.nn.init.normal_(tensor, mean=0.0, std=1.0)

• tensor 是输入张量
• mean 是正态分布均值
• std 是正态分布标准差

参考链接

1. 从零开始行人重识别
2. Person_reID_baseline_pytorch QA
3. pytorch 文档 torch.nn.init.kaiming_normal_
4. pytorch 文档 torch.nn.Module.apply
5. torch.nn.Module.apply(fn)
6. 【torch杂记】torch.nn.init.kaiming_normal_
7. 神经网络中的权重初始化一览：从基础到Kaiming