手写数字数据集AutoEncoder降噪算法

对训练数据加噪声的方法，在训练里面对 x 做如下处理，添加椒盐噪声：

        bs, ch, h, w = x.shape
        x = x.reshape(bs, ch, h*w) + 0.2*np.random.normal(size=28*28)
        x = x.to(torch.float32)

数据集里面的标签 label 无用，因为 AutoEncoder 去噪是无监督方法。

一、读取数据

import torch
import torch.nn as nn
import torchvision
import torch.optim as optim
import matplotlib.pyplot as plt
import numpy as np

EPOCH = 5
BATCH_SIZE = 64
LR = 0.001
DOWNLOAD_MNIST = True
N_TEST_IMG = 5

train_data = torchvision.datasets.MNIST(
    root='../mnist_data/',
    train=True,
    transform=torchvision.transforms.ToTensor(),
    download=DOWNLOAD_MNIST,
    )

test_data = torchvision.datasets.MNIST(
    root='../mnist_data/',
    train=False,
    transform=torchvision.transforms.ToTensor(),
    download=DOWNLOAD_MNIST,
    )
print(train_data.train_data.size())
print(train_data.train_labels.size())

train_loader=Data.DataLoader(dataset=train_data, batch_size=BATCH_SIZE,shuffle=True)
test_loader=Data.DataLoader(dataset=test_data, batch_size=BATCH_SIZE,shuffle=False)

二、前3步：构建模型、设置优化器、损失函数

class AutoEncoder(nn.Module):
    def __init__(self):
        super(AutoEncoder, self).__init__()
        self.encoder = nn.Sequential(
        nn.Linear(28*28, 128),
        nn.ReLU(),
        nn.Linear(128, 64),
        nn.ReLU(),
        nn.Linear(64, 12),
        nn.ReLU(),
        #             nn.Linear(12, 3),
        )
        self.decoder = nn.Sequential(
        #             nn.Linear(3, 12),
        #             nn.Tanh(),
            nn.Linear(12, 64),
            nn.ReLU(),
            nn.Linear(64, 128),
            nn.ReLU(),
            nn.Linear(128, 28*28),
        #             nn.Sigmoid(),
        )
    def forward(self, x):
        encoded = self.encoder(x)
        decoded = self.decoder(encoded)
        return decoded
AE = AutoEncoder()
optimizer = optim.Adam(AE.parameters(), lr=LR)
loss_func = nn.MSELoss()
# 1 2 3

三、后5步：前向计算、计算损失、no_grad, backward, step，如果有验证集的话，每到一定步数在no_grad下进行验证，不需要zer_grad和backward

for epoch in range(EPOCH):
    for step, (x, _) in enumerate(train_loader):
        bs, ch, h, w = x.shape
        x = x.reshape(bs, ch, h*w) + 0.2*np.random.normal(size=28*28)
        x = x.to(torch.float32)
        # 4 5
        code = AE.encoder(x)  # https://blog.csdn.net/weixin_55191433/article/details/121402942
        recon = AE.decoder(code)
        loss = loss_func(recon, x)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        
        if step % 100 == 0:
            print('Epoch:', epoch, ' | train loss: %.4f'%loss.item())

四、查看结果（测试集）

cnt = 16
idx = 0
plt.figure(1)
with torch.no_grad():
    for step, (x, _) in enumerate(test_loader):
        bs, ch, h, w = x.shape
        x = x.reshape(bs, ch, h*w) + 0.2*np.random.normal(size=28*28)
        x = x.to(torch.float32)
        # 4 5
        code = AE.encoder(x)  # https://blog.csdn.net/weixin_55191433/article/details/121402942
        recon = AE.decoder(code)
        print(recon.shape)
        for i in range(16):
            plt.subplot(4,4,step+1)
            img = recon[i].squeeze().reshape(28, 28)
            plt.imshow(img)
        # loss = loss_func(recon, x)
        idx += 1
        if idx == 16:
            break

结果如下：加噪声后，和通过AE去噪后。