Question

我有（100，64，64，3）形状的输入数据（彩色图像），并尝试训练带有2个conv / pooling层的CNN进行二进制分类。我一直遇到尺寸不匹配错误。还尝试将图像重塑为（-1、3、64、64）大小

class SimpleCNN(nn.Module):

    def __init__(self, input_dim, hidden_dim, output_dim, kernel_size):
        super(SimpleCNN, self).__init__()

        self.conv1 = nn.Conv2d(3, 10, kernel_size, padding=0)
        self.conv2 = nn.Conv2d(10, 20, kernel_size, padding=0)
        self.fc1 = nn.Linear(hidden_dim*16*16, hidden_dim)
        self.fc2 = nn.Linear(output_dim, output_dim)

    def forward(self, x):
        x = F.max_pool2d((self.conv1(x)), 2)
        x = F.max_pool2d(F.relu(self.conv2(x)), 2)
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        return F.softmax(x)

net = SimpleCNN(in_channels, hidden_dim, out_channels, kernel_size)

criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(net.parameters(), lr=learning_rate)

losses = []
idx = []
count=1
for epoch in range(num_epochs):
    for i, (images, labels) in enumerate(train_loader):
        images = Variable(images.view(-1, 3, n_pixel, n_pixel))
        labels = Variable(labels)

        # Intialize the hidden weight to all zeros
        optimizer.zero_grad()
        # Forward pass: compute the output class given a image
        outputs = net(images)
        # Compute the loss: difference between the output class and the pre-given label
        loss = criterion(outputs, labels)
        # Backward pass: compute the weight
        loss.backward()
        # Optimizer: update the weights of hidden nodes
        optimizer.step()
        count+=1

        if (i+1) % 10 == 0:
            idx.append(count)
            losses.append(loss.data.numpy().tolist())

完整错误消息为

仅支持批次的空间目标（3D张量），但尺寸目标为：/pytorch/aten/src/THNN/generic/SpatialClassNLLCriterion.c:61'

如何解决pyTorch中图像上CNN的尺寸不匹配错误？

0 个答案: