sakigake3421 / demosaic_project

Removing Pixelated Mosaic Censorship using ESRGAN and green_mask_project

demosaic_project

Removing Pixelated Mosaic Censorship using ESRGAN (Enhanced SRGAN) and green_mask_project

ESRGAN model should be downloaded here: Twittman's 4x_FatalPixels_340000_G.pth, and placed into models folder!

To have a decent performance you need a CUDA+cudnn(NVIDIA) or ROCm(AMD) compatible GPU with installed drivers! Otherweis use "--cpu" key, but beware it can be VERY slow.

HOW TO USE:

1. Place your images inside decensor_input
2. Run "python demosaic_project_ESRGAN.py" or ("python demosaic_project_ESRGAN.py --cpu" for cpu version)
3. Take your decensored images from decensor_output
4. ???
5. Profit

Don't forget to install python3 and all the requirements with command "pip3 install -r requirements.txt" in cmd or bash

TODO:

• GUI
• Config file
• Better detection algorithm

Inspiration from DeepCreamPy and hent-AI
Credits to Twittman for making trained model Fatal Pixels.

Sample image by hks(@timbougami)