... F.relu(x) x = F.max_pool2d(x, 2) x = self.dropout(x) x = torch.flatten(x, ... 要predict),所以用flatten 把feature 集中成vector 後,再用fully-connected ...
確定! 回上一頁