python 中的 CNN 模块给出错误大小不匹配，m1：[12288 x 26]，m2：[12288 x 26]

Question

I'm having some issues with my CNN model and I don't understand what I'm doing wrong. I've tried to change my model multiple times to look like m1: 12288 x 26 and 26 x 12288, but I'm not quite sure how to do that. Could anyone please help me out?

 self.conv1 = torch.nn.Conv1d(input_size, 256, kernel_size)
 self.conv2 = torch.nn.Conv1d(256, 256, kernel_size)
 self.fc1 = torch.nn.Linear(256*input_size, output_size)

or

 #self.conv1 = torch.nn.Conv1d(48, 256, 1)
 #self.conv2 = torch.nn.Conv1d(256, 48, 1)
 #self.fc1 = torch.nn.Linear(48*256, 26)

The model looks like:

CNN(
(conv1): Conv1d(48, 256, kernel_size=(1,), stride=(1,))
(conv2): Conv1d(256, 256, kernel_size=(1,), stride=(1,))
(fc1): Linear(in_features=12288, out_features=26, bias=True)
)

The error I get "RuntimeError: size mismatch, m1: [12288 x 26], m2: [12288 x 26] at..\aten\src\TH/generic/THTensorMath.cpp:41"

Answer 1

Need to "flatten" it down: self.fc1 = torch.nn.Linear(256(48-(kernel_size-1)2), output_size) did the trick.