agent-inf / inpainting

Another Implementation of "Context Encoders: Feature Learning by Inpainting"

inpainting

Another Implementation of "Context Encoders: Feature Learning by Inpainting"

Use convert_image.py to convert all jpg images in a folder into one or more binary files, which will use for training or validating or testing.

Then run "python inpaint.py" and the training will started.

The program will save sample images every epoch in ./samples folder by default.

I run my program in Ubuntu 16.04, Tensorflow 1.2 and python 2.7.

PS: I've made lots of changes compare to jazzsaxmafia/Inpainting because this is part of my private project, but it will do the inpainting job.

inpainting's Issues

block location of the picture

Thank you for sharing your code ！！！I appreciate it ！！！
Recently, I have studied your code, I have a problem , in the ' inpaint.py ' when I choose the 'center' from the '(center, random)', I want to block the center of the picture.
But in the sample folder , I find the picture, block location is random , I don' t know why

About the number '127.5' in this code

Hello,I have a question .
I know the size of images is 128 *128 in this code.
In this code, the 194th line of inpaint.py : image_batch = (image_batch.astype(np.float32) - 127.5) / 127.5
and the141th line of utils.py : image = image * 127.5 + 127.5.
So, I want to know , is the number '127.5' associated to the size of image? And if the size of my own images is 256 *360, shall I change the number '127.5'?
I could appreciate it you can give me some advices.
Thank you ！！！

Can you show some results of you implementation.

Thanks.
And are irregular holes supported?