RuntimeError：尺寸不匹配，m1：[32 x 1]，m2：[32 x 9]

Question

我正在构建一个 CNN 并对其进行字母 A 到 I（9 类）的手势分类训练，每个图像都是 224x224 大小的 RGB。

不确定我需要转置哪个矩阵以及如何转置。我已经设法匹配层的输入和输出，但是矩阵乘法的事情，不太确定如何解决它。

class LargeNet(nn.Module):

    def __init__(self):

        super(LargeNet, self).__init__()

        self.name = "large"

        self.conv1 = nn.Conv2d(3, 5, 5)

        self.pool = nn.MaxPool2d(2, 2)

        self.conv2 = nn.Conv2d(5, 10, 5)

        self.fc1 = nn.Linear(10 * 53 * 53, 32)

        self.fc2 = nn.Linear(32, 9)

    def forward(self, x):

        x = self.pool(F.relu(self.conv1(x)))

        print('x1')

        x = self.pool(F.relu(self.conv2(x)))

        print('x2')

        x = x.view(-1, 10*53*53)

        print('x3')

        x = F.relu(self.fc1(x))

        print('x4')

        x = x.view(-1, 1)

        x = self.fc2(x)

        print('x5')

        x = x.squeeze(1) # Flatten to [batch_size]

        return x

和培训代码

#Loss and optimizer

criterion = nn.BCEWithLogitsLoss()

optimizer = optim.SGD(model2.parameters(), lr=learning_rate, momentum=0.9)

# Train the model

total_step = len(train_loader)

loss_list = []

acc_list = []

for epoch in range(num_epochs):

    for i, (images, labels) in enumerate(train_loader):

        print(i,images.size(),labels.size())

        # Run the forward pass

        outputs = model2(images)

        labels=labels.unsqueeze(1)

        labels=labels.float()

        loss = criterion(outputs, labels)

代码打印到 x4，然后我收到此错误 RuntimeError: size mismatch, m1: [32 x 1], m2: [32 x 9] at C:\w\1\s\tmp_conda_3.7_055457\conda\conda- bld\pytorch_1565416617654\work\aten\src\TH/generic/THTensorMath.cpp:752

完整的回溯错误：https ://ibb.co/ykqy5wM

Answer 1

您不需要x=x.view(-1,1)andx = x.squeeze(1)在您的forward功能中。删除这两行。您的输出形状将是(batch_size, 9).此外，您需要转换labels为 one-hot 编码，其形状为(batch_size, 9).class LargeNet(nn.Module):    def __init__(self):        super(LargeNet, self).__init__()        self.name = "large"        self.conv1 = nn.Conv2d(3, 5, 5)        self.pool = nn.MaxPool2d(2, 2)        self.conv2 = nn.Conv2d(5, 10, 5)        self.fc1 = nn.Linear(10 * 53 * 53, 32)        self.fc2 = nn.Linear(32, 9)    def forward(self, x):        x = self.pool(F.relu(self.conv1(x)))        x = self.pool(F.relu(self.conv2(x)))        x = x.view(-1, 10*53*53)        x = F.relu(self.fc1(x))        x = self.fc2(x)        return xmodel2 = LargeNet()#Loss and optimizercriterion = nn.BCEWithLogitsLoss()# nn.BCELoss()optimizer = optim.SGD(model2.parameters(), lr=0.1, momentum=0.9)images = torch.from_numpy(np.random.randn(2,3,224,224)).float() # fake images, batch_size is 2labels = torch.tensor([1,2]).long() # fake labelsoutputs = model2(images)one_hot_labels = torch.eye(9)[labels] loss = criterion(outputs, one_hot_labels)