pytorch 用F.normalization的逆归一化如何操作

逆归一化的时候再把这个数乘回去就行了

magnitude = a.norm(p=2, dim=1, keepdim=True) # NEW

a=torch.nn.functional.normalize(a, p=2, dim=1)

a_or= a* magnitude # NEW

print(a_or)

Outputs:

tensor(\[\]1,2,3)