torch3652

channel error

2

 If you add another layer before the last loop so that the final output is 513 channels, but the last convolution accepts 512 channels, there is an...

您好，源代码中的/models/transformer_cosine.py部分的第66行，这里feature的特征为什么是(64,3,512)的形状，这样无法与mask做矩阵乘法。