Giters

naoto0804 / pytorch-inpainting-with-partial-conv

Unofficial pytorch implementation of 'Image Inpainting for Irregular Holes Using Partial Convolutions' [Liu+, ECCV2018]

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The input image should be a square with equal height and weight? How to keep the output image the same size as input?

jobsfan opened this issue · comments

commented

Hi @naoto0804 , I finally got the code running! Thank you for sharing.

I have two problems in using it.

  1. how to break the rule for the setted square size? I mean that I have to input a image with a 512x512, the same height and width. Is it possible to input any size of image?

  2. the output image are so small grid, Is it possible to output a image the same size with the input?

commented
  1. You can test an arbitrary size since the network is fully-convolutional. During testing, I recommend you to resize the image being closer to 256x256 before feeding it to the model, since the model is trained on that resolution.
  2. Please see evaluation.py to handle the estimated outputs. I just concatenated them to make grids.