Hi everyone
I wanna resize some variant-sized Variables into a uniform size without losing their grad info
For instance: (1 x 3 x W1 x H1) (1 x 3 x W2 x H2) etc. —> (1 x 3 x W x H)
I found this may help, but I had no idea how to make it work.
Any suggestions will be appreciated ! Thanks
PS: if I transform those variant-sized tensors into images and apply torchvision.transforms.Resize to them, does it mean that I will not be able to use their gradients any more ?
Solved: torch.nn.functional.upsample with bilinear works for me