Upsample Nearest Implementation
1. 通过Reshape和Concat结合实现Upsample
由于一些框架，不支持Upsample，可以用ReShape和Concat这种常见的Op替换
class MyUpSample(nn.Module):
dump_patches = True
def __init__(self, scale_factor = 2):
super(MyUpSample, self).__init__()
#self._upsample = nn.Upsample(scale_factor=scale_factor)
def forward(self, x):
#return self._upsample(x)
ss = x.size()
x = x.view(ss[0], ss[1], ss[2] * ss[3], 1) # 1
ups1 = torch.cat((x, x), 3) # 2
ups1 = ups1.view(ss[0], ss[1], ss[2], 2 * ss[3]) # 3
ups2 = torch.cat((ups1, ups1), 3) # 4
ups2 = ups2.view(ss[0], ss[1], ss[2] * 2, ss[3] * 2) # 5
return ups2其中代码注释中的5个步骤，分别对应下图中的5步：
