# singanInpaint

**Repository Path**: zyabo/singan-inpaint

## Basic Information

- **Project Name**: singanInpaint
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-12-29
- **Last Updated**: 2026-07-15

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# SinGAN extension: application to inpainting
[Original paper Arxiv](https://arxiv.org/pdf/1905.01164.pdf) 

Forks https://github.com/tamarott/SinGAN

This repo essentially adds a `recover.py` file which should be used for inpainting.


Example 
```
python recover.py --input_name birds.png --fake_input_name birds_bis.png --disc_loss 0.01 --use_mask True
```

`input_name` is the file name of the uncorrupted image to train the model on (we assume that we have such an image).

`fake_input_name` is the file name of another image which has the same multi-level patch distribution. If `use_mask=True`, 
a mask is added to produce a corrupted image. Optional parameters allow you to control the mask size and position.