基于Pytorch的特征图提取

简述

为了方便理解卷积神经网络的运行过程，需要对卷积神经网络的运行结果进行可视化的展示。

大致可分为如下步骤：

• 单个图片的提取
• 神经网络的构建
• 特征图的提取
• 可视化展示

单个图片的提取

根据目标要求，需要对单个图片进行卷积运算，但是Pytorch中读取数据主要用到torch.utils.data.DataLoader类，因此我们需要编写单个图片的读取程序

def get_picture(picture_dir, transform):
    '''
    该算法实现了读取图片，并将其类型转化为Tensor
    '''
    tmp = []
    img = skimage.io.imread(picture_dir)
    tmp.append(img)
    img = skimage.io.imread('./picture/4.jpg')
    tmp.append(img)
    img256 = [skimage.transform.resize(img, (256, 256)) for img in tmp]
    img256 = np.asarray(img256)
    img256 = img256.astype(np.float32)

    return transform(img256[0])

注意： 神经网络的输入是四维形式，我们返回的图片是三维形式，需要使用unsqueeze()插入一个维度

神经网络的构建

网络的基于LeNet构建，不过为了方便展示，将其中的参数按照2562563进行的参数的修正

网络构建如下：

class LeNet(nn.Module):
    '''
    该类继承了torch.nn.Modul类
    构建LeNet神经网络模型
    '''
    def __init__(self):
        super(LeNet, self).__init__()

        # 第一层神经网络，包括卷积层、线性激活函数、池化层
        self.conv1 = nn.Sequential( 
            nn.Conv2d(3, 32, 5, 1, 2),   # input_size=(3*256*256)，padding=2
            nn.ReLU(),                  # input_size=(32*256*256)
            nn.MaxPool2d(kernel_size=2, stride=2),  # output_size=(32*128*128)
        )

        # 第二层神经网络，包括卷积层、线性激活函数、池化层
        self.conv2 = nn.Sequential(
            nn.Conv2d(32, 64, 5, 1, 2),  # input_size=(32*128*128)
            nn.ReLU(),            # input_size=(64*128*128)
            nn.MaxPool2d(2, 2)    # output_size=(64*64*64)
        )

        # 全连接层(将神经网络的神经元的多维输出转化为一维)
        self.fc1 = nn.Sequential(
            nn.Linear(64 * 64 * 64, 128),  # 进行线性变换
            nn.ReLU()                    # 进行ReLu激活
        )

        # 输出层(将全连接层的一维输出进行处理)
        self.fc2 = nn.Sequential(
            nn.Linear(128, 84),
            nn.ReLU()
        )

        # 将输出层的数据进行分类(输出预测值)
        self.fc3 = nn.Linear(84, 62)

    # 定义前向传播过程，输入为x
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        # nn.Linear()的输入输出都是维度为一的值，所以要把多维度的tensor展平成一维
        x = x.view(x.size()[0], -1)
        x = self.fc1(x)
        x = self.fc2(x)
        x = self.fc3(x)
        return x

特征图的提取

直接上代码：

class FeatureExtractor(nn.Module):
    def __init__(self, submodule, extracted_layers):
        super(FeatureExtractor, self).__init__()
        self.submodule = submodule
        self.extracted_layers = extracted_layers
 
    def forward(self, x):
        outputs = []
        for name, module in self.submodule._modules.items():
        # 目前不展示全连接层
            if "fc" in name: 
                x = x.view(x.size(0), -1)
            print(module)
            x = module(x)
            print(name)
            if name in self.extracted_layers:
                outputs.append(x)
        return outputs

可视化展示

可视化展示使用matplotlib

代码如下：

    # 特征输出可视化
    for i in range(32):
        ax = plt.subplot(6, 6, i + 1)
        ax.set_title('Feature {}'.format(i))
        ax.axis('off')
        plt.imshow(x[0].data.numpy()[0,i,:,:],cmap='jet')
    plt.plot()

完整代码

在此贴上完整代码

import os
import torch
import torchvision as tv
import torchvision.transforms as transforms
import torch.nn as nn
import torch.optim as optim
import argparse
import skimage.data
import skimage.io
import skimage.transform
import numpy as np
import matplotlib.pyplot as plt

# 定义是否使用GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load training and testing datasets.
pic_dir = './picture/3.jpg'

# 定义数据预处理方式(将输入的类似numpy中arrary形式的数据转化为pytorch中的张量（tensor）)
transform = transforms.ToTensor()


def get_picture(picture_dir, transform):
    '''
    该算法实现了读取图片，并将其类型转化为Tensor
    '''
    img = skimage.io.imread(picture_dir)
    img256 = skimage.transform.resize(img, (256, 256))
    img256 = np.asarray(img256)
    img256 = img256.astype(np.float32)

    return transform(img256)


def get_picture_rgb(picture_dir):
    '''
    该函数实现了显示图片的RGB三通道颜色
    '''
    img = skimage.io.imread(picture_dir)
    img256 = skimage.transform.resize(img, (256, 256))
    skimage.io.imsave('./picture/4.jpg',img256)

    # 取单一通道值显示
    # for i in range(3):
    #     img = img256[:,:,i]
    #     ax = plt.subplot(1, 3, i + 1)
    #     ax.set_title('Feature {}'.format(i))
    #     ax.axis('off')
    #     plt.imshow(img)

    # r = img256.copy()
    # r[:,:,0:2]=0
    # ax = plt.subplot(1, 4, 1)
    # ax.set_title('B Channel')
    # # ax.axis('off')
    # plt.imshow(r)

    # g = img256.copy()
    # g[:,:,0]=0
    # g[:,:,2]=0
    # ax = plt.subplot(1, 4, 2)
    # ax.set_title('G Channel')
    # # ax.axis('off')
    # plt.imshow(g)

    # b = img256.copy()
    # b[:,:,1:3]=0
    # ax = plt.subplot(1, 4, 3)
    # ax.set_title('R Channel')
    # # ax.axis('off')
    # plt.imshow(b)

    # img = img256.copy()
    # ax = plt.subplot(1, 4, 4)
    # ax.set_title('image')
    # # ax.axis('off')
    # plt.imshow(img)

    img = img256.copy()
    ax = plt.subplot()
    ax.set_title('image')
    # ax.axis('off')
    plt.imshow(img)

    plt.show()


class LeNet(nn.Module):
    '''
    该类继承了torch.nn.Modul类
    构建LeNet神经网络模型
    '''
    def __init__(self):
        super(LeNet, self).__init__()

        # 第一层神经网络，包括卷积层、线性激活函数、池化层
        self.conv1 = nn.Sequential( 
            nn.Conv2d(3, 32, 5, 1, 2),   # input_size=(3*256*256)，padding=2
            nn.ReLU(),                  # input_size=(32*256*256)
            nn.MaxPool2d(kernel_size=2, stride=2),  # output_size=(32*128*128)
        )

        # 第二层神经网络，包括卷积层、线性激活函数、池化层
        self.conv2 = nn.Sequential(
            nn.Conv2d(32, 64, 5, 1, 2),  # input_size=(32*128*128)
            nn.ReLU(),            # input_size=(64*128*128)
            nn.MaxPool2d(2, 2)    # output_size=(64*64*64)
        )

        # 全连接层(将神经网络的神经元的多维输出转化为一维)
        self.fc1 = nn.Sequential(
            nn.Linear(64 * 64 * 64, 128),  # 进行线性变换
            nn.ReLU()                    # 进行ReLu激活
        )

        # 输出层(将全连接层的一维输出进行处理)
        self.fc2 = nn.Sequential(
            nn.Linear(128, 84),
            nn.ReLU()
        )

        # 将输出层的数据进行分类(输出预测值)
        self.fc3 = nn.Linear(84, 62)

    # 定义前向传播过程，输入为x
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        # nn.Linear()的输入输出都是维度为一的值，所以要把多维度的tensor展平成一维
        x = x.view(x.size()[0], -1)
        x = self.fc1(x)
        x = self.fc2(x)
        x = self.fc3(x)
        return x

# 中间特征提取
class FeatureExtractor(nn.Module):
    def __init__(self, submodule, extracted_layers):
        super(FeatureExtractor, self).__init__()
        self.submodule = submodule
        self.extracted_layers = extracted_layers
 
    def forward(self, x):
        outputs = []
        print(self.submodule._modules.items())
        for name, module in self.submodule._modules.items():
            if "fc" in name: 
                print(name)
                x = x.view(x.size(0), -1)
            print(module)
            x = module(x)
            print(name)
            if name in self.extracted_layers:
                outputs.append(x)
        return outputs


def get_feature():
    # 输入数据
    img = get_picture(pic_dir, transform)
    # 插入维度
    img = img.unsqueeze(0)

    img = img.to(device)

    # 特征输出
    net = LeNet().to(device)
    # net.load_state_dict(torch.load('./model/net_050.pth'))
    exact_list = ["conv1"，"conv2"]
    myexactor = FeatureExtractor(net, exact_list)
    x = myexactor(img)

    # 特征输出可视化
    for i in range(32):
        ax = plt.subplot(6, 6, i + 1)
        ax.set_title('Feature {}'.format(i))
        ax.axis('off')
        plt.imshow(x[0].data.numpy()[0,i,:,:],cmap='jet')

    plt.show()

# 训练
if __name__ == "__main__":
    get_picture_rgb(pic_dir)
    # get_feature()