Pytorch: In Con2d layer,how to pad "Same" not "Zero"

Created on 17 Mar 2017  ·  3Comments  ·  Source: pytorch/pytorch

In Con2d layer,how to pad "Same" not "Zero".

In tensorflow , i see the option to choose whether pad the "Same" pixel as edge or pad "Zero" pixel .But dose pytorch have that option?

picture a1363901216

Most helpful comment

pytorch does not have this option. if you want to do something like this, you have to explicitly pad via nn.functional.pad in the replicate mode: http://pytorch.org/docs/nn.html#torch.nn.functional.pad

picture soumith on 17 Mar 2017
👍3

All 3 comments

pytorch does not have this option. if you want to do something like this, you have to explicitly pad via nn.functional.pad in the replicate mode: http://pytorch.org/docs/nn.html#torch.nn.functional.pad

soumith picture soumith on 17 Mar 2017
👍3

thanks

a1363901216 picture a1363901216 on 17 Mar 2017

Why does pytorch not have this feature? It would greatly help with quick prototyping

rtaori picture rtaori on 4 Jul 2019
Was this page helpful?
0 / 5 - 0 ratings

Related issues

negrinho picture negrinho  ·  3Comments
miguelvr picture miguelvr  ·  3Comments
keskarnitish picture keskarnitish  ·  3Comments
soumith picture soumith  ·  3Comments
ikostrikov picture ikostrikov  ·  3Comments