https://github.com/mattyamonaca/PBRemTools

Precise background remover
https://github.com/mattyamonaca/PBRemTools

Last synced: about 2 months ago
JSON representation

Precise background remover

• Host: GitHub
• URL: https://github.com/mattyamonaca/PBRemTools
• Owner: mattyamonaca
• License: mit
• Created: 2023-04-04T23:38:33.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-11-16T19:21:20.000Z (10 months ago)
• Last Synced: 2024-07-09T02:02:49.066Z (3 months ago)
• Language: Python
• Size: 57.6 KB
• Stars: 338
• Watchers: 5
• Forks: 24
• Open Issues: 23
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Latest Updates

Combine Segment Anything and CLIP and control the generating mask with text has been added!

![スクリーンショット 2023-04-11 115936](https://user-images.githubusercontent.com/48423148/231046410-3322ad98-f972-40cf-b51d-84ff3e91cb2a.png)

Segmentation can also be performed from the PBRemTools tab.

However, to use it, the Segment Anything Model must be added manually.

Download the model from here (https://github.com/facebookresearch/segment-anything#model-checkpoints) and place it under "models"(stable-diffusion-webui/extensions/PBRemTools/models).

This program is prepared with reference to (https://github.com/facebookresearch/segment-anything) and (https://github.com/Curt-Park/segment-anything-with-clip)

# PBRemTools

PBRemTools(Precise background remover tools) is a collection of tools to crop backgrounds from a single picture with high accuracy.

- Base Image



- PBRemTools(Tile division ABG Remover)



- PBRemTools(CascadePSP)



- ABG Remover



- RemBG



# Tools

## Tile division ABG Remover

This tool is based on Anime Remove Background(https://huggingface.co/spaces/skytnt/anime-remove-background) and ABG_extension(https://github.com/KutsuyaYuki/ABG_extension).

Post-processing is added for more precise cropping based on the mask image generated by Anime Remove Background.

In this post-processing step, the input image is divided into a specified number of tiles, and the pixels in each tile are clustered based on color information.



Extract clusters whose mask image content exceeds a threshold value as foreground.

### Parameters

- horizontal split num: Number of horizontal tile divisions.

- vertical split num: Number of vertical tile segments.

- n_cluster: Number of clusters based on color information.

- alpha threshold: Transparency of the mask considered as foreground.

- mask content ratio: Threshold for how much mask a cluster should contain to be considered foreground.

## CascadePSP

This tool is based on CascadePSP(https://github.com/hkchengrex/CascadePSP).

# Installation

- Stable diffusion web ui.

Install from webui's Extensions tab.

## Precautions

This program uses code that contains the Apache License 2.0

# API

Currently, there are two APIs available:

 - To get the currently available SAM model: `GET http://localhost:7861/pbrem/sam-model`

 - To process an image: `POST http://localhost:7861/pbrem/predict`

After launching the web UI API, you can visit the detailed API documentation at `http://localhost:7861/redoc`