Skip to content
This repository has been archived by the owner on May 12, 2022. It is now read-only.

StevenCyb/Inpainting-using-Context-Encoder-in-Generative-Adversarial-Network

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Inpainting using Context-Encoder and Generative-Adversarial-Network

I was excited about the inpainting project from here, which in turn is based on the paper Context Encoders: Feature Learning by Inpainting. Therefore I wanted to try my own implementation. Some parts are from the other project, so have a look at the other project and the original paper. In addition, I also added some own modifications, e.g. an additional layer in the decoder or the tiling of the pictures.

Table Of Contents

Small Evaluation

The following result needed 1:10h with batch size of 1 and 30k epochs on a GTX1080TI. Training_vis results

How To Use It

First I created the directory weights. Then I copied the training data into the directory training. For the upper training I used the following command:

python3 train.py -mar 3 -mal 0 -mac 0 -bs 1

The arguments -mar define the max. ammount of random ractangles for the random mask. In addition I disable the random drawing of lines and circles by setting -mal and -mac to 0. But you can play around.

Then I run the following command to perform a prediction:

python3 predict.py -i ./media/0.jpg -m ./media/0_mask.jpg -o ./media/0_prediction.jpg

The Input images and masks looked like the Masked-Image and Mask from the upper figure. Where the white area characterised the region to inpaint.

You can get more information about available arguments by using the argument -h.

Releases

No releases published

Packages

No packages published

Languages