How to change the size of Variable.data in forward function?

1

I want to add a module in the network, the module is used to resize the size of 4-D tensor in the network, Do you know how to achieve it?

2

It seems like you haven’t finish the 60 min tutorials

def forward(self, x):
       x = x.view(....) # I think you mean reshape instead of resize
1 Like
3

I’m sorry that I don’t describe the problem clearly. I want to crop or resize the size of variable.data in the forward function.

4

resize op can not backpropagate.
do you mean index like variable[:100, :10]

5

Yes,I just want to crop the Variable.data (such as 258258 -> 256256), is it feasible?

6

You can not do it directly in variable.data, It will raise error of size-dismatch when backward.
Try variable[1:257,1:257] if you know what you are doing.

1 Like
7

Thank you very much! I have solved the problem in this way.

8

By the way, I want to know that whether the tensor’s ops can be used for Variable, such as torch.add(), torch.sigmoid() and so on .

9

torch.add is overwrite in variable which made it an op on variable. So it is ok to use torch.func on variable in most cases.