请慎用torch.nn.Upsample

版本：torch==1.12.1

计算机视觉算法中的上采用函数，一般都是采用nearest这种简单模式进行，PyTorch也是有相应的模块`nn.Upsample`来支持。但这个模块实在是不太与时俱进，慎用！

1. 不支持bf16

RuntimeError: "upsample_nearest2d_out_frame" not implemented for 'BFloat16'

2. nn.Upsample 导致模型可复现性变差

参考这里。

---

解法就是，自定义Upsample模块，这里也是参考了这里：

import torch.nn as nn
class UpsampleDeterministic(nn.Module):
    def __init__(self,upscale=2):
        super(UpsampleDeterministic, self).__init__()
        self.upscale = upscale

    def forward(self, x):
        '''
        x: 4-dim tensor. shape is (batch,channel,h,w)
        output: 4-dim tensor. shape is (batch,channel,self.upscale*h,self.upscale*w)
        '''
        return x[:, :, :, None, :, None]\
        .expand(-1, -1, -1, self.upscale, -1, self.upscale)\
        .reshape(x.size(0), x.size(1), x.size(2)\
                 *self.upscale, x.size(3)*self.upscale)