如何解决MNIST 1D CNN训练中Conv1d输入维度不匹配的RuntimeError

我因研究需求必须使用nn.Conv1d构建1D CNN训练MNIST，但运行时触发了输入维度错误。我明确需要将输入从[96,1,28,28]转换为[96,1,784]的形状以适配Conv1D的要求，但不清楚应该在代码的哪个阶段（数据加载/模型forward）实现，之前尝试过ChatGPT给出的方案但无效，也没找到合适的解决方法。

现有代码 #

模型定义 #

import torch.nn as nn

class net_mnist(nn.Module):
    def __init__(self, input_size, output_size):
        super(net_mnist, self).__init__()
        self.conv1 = nn.Conv1d(1, 1, kernel_size=1, stride=1, padding=1)
        self.pool = nn.MaxPool1d(kernel_size=1, stride=2)
        self.fc1 = nn.Linear(input_size, 4096, bias=True)
        self.fc2 = nn.Linear(4096, 4096, bias=True)
        self.fc3 = nn.Linear(4096, 4096, bias=True)
        self.fc4 = nn.Linear(4096, 4096, bias=True)
        self.fc5 = nn.Linear(4096, output_size, bias=True)
        self.relu = nn.ReLU()

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.pool(x)
        x = self.conv2(x)  # 注意：conv2未在__init__中定义，会触发AttributeError
        x = self.relu(x)
        x = self.pool(x)
        x = x.view(x.size(0), -1)
        x = self.fc1(x)
        x = self.relu(x)
        x = self.fc2(x)
        return x
        return x.squeeze()  # 这段代码永远不会执行

数据集加载代码 #

from torchvision.transforms import Compose, ToTensor
from torchvision.datasets import MNIST
from torch.utils.data import DataLoader

transforms = Compose([
    ToTensor(),
    # Normalize(
    #     mean=[0.1307],
    #     std=[0.3081],
    # )
])
trainset = MNIST(root='./mnist_data', train=True, download=True, transform=transforms)
testset = MNIST(root='./mnist_data', train=False, download=True, transform=transforms)

trainloader = DataLoader(trainset, batch_size=96, shuffle=True)
testloader = DataLoader(testset, batch_size=96, shuffle=True)

训练执行代码 #

net = net_mnist(28*28, 10)
net.to(device)
criterion = nn.CrossEntropyLoss().to(device)
writer = SummaryWriter()
net = train_mnist(net, trainloader, testloader, criterion, lrate=0.03, max_epochs=6)

错误信息 #

RuntimeError                              Traceback (most recent call last)
    <ipython-input-21-6bc9f6dcde7f> in <cell line: 5>()
      3 criterion = nn.CrossEntropyLoss().to(device)
      4 writer = SummaryWriter()
----> 5 net = train_mnist(net,trainloader,testloader,criterion,lrate=0.03,max_epochs=6)

5 frames
    /usr/local/lib/python3.10/dist-packages/torch/nn/modules/conv.py in _conv_forward(self, input, weight, bias)
    307                             weight, bias, self.stride,
    308                             _single(0), self.dilation, self.groups)
---> 309         return F.conv1d(input, weight, bias, self.stride,
    310                         self.padding, self.dilation, self.groups)
    311 

RuntimeError: Expected 2D (unbatched) or 3D (batched) input to conv1d, but got input of size: [96, 1, 28, 28]

核心需求 #

1. 将MNIST输入从[batch_size, 1, 28, 28]转换为[batch_size, 1, 784]，适配nn.Conv1d的输入要求（3D张量：[batch_size, in_channels, sequence_length]）
2. 修复模型中未定义conv2的问题
3. 确保模型forward流程正确（当前末尾的return x.squeeze()永远不会执行）

内容的提问来源于stack exchange，提问作者jerry gill