Loss is always around 5.6 and doesn't decrease, Accuracy is at 0.64% - 0.84%

[vdw (Chris)]

@Ben_Sin

batch_size, seq_length, channels, height, width = x.size()
x = x.reshape(batch_size, channels, seq_length, height, width)

You don’t want to use reshape() to swap dimensions. This will mess up your tensors; more details here. You probably want:

x = x.transpose(1, 2)