因为用到了BatchNorm,所以batch_size要大于1,drop_last 参数设置为True。
如果是两个GPU训练,batch_size改为4。
参考
ValueError: Expected more than 1 value per channel when training, got input size [1, 16, 1, 1](解决方案)
Pytorch遇到报错Expected more than 1 value per channel when training, got input size torch.Size