README.md

---
library_name: matanyone2
tags:
- model_hub_mixin
- pytorch_model_hub_mixin
---

<div align="center">
<img src="https://github.com/pq-yang/MatAnyone2/blob/main/assets/matanyone2_logo.png?raw=true" alt="MatAnyone Logo" style="height: 52px;">
<div style="text-align: center">
    <h2>Scaling Video Matting via a Learned Quality Evaluator</h2>
</div>

<div>
    <a href='https://pq-yang.github.io/' target='_blank'>Peiqing Yang</a><sup>1</sup>&emsp;
    <a href='https://shangchenzhou.com/' target='_blank'>Shangchen Zhou</a><sup>1†</sup>&emsp;
    <a href="https://www.linkedin.com/in/kai-hao-794321382/" target='_blank'>Kai Hao</a><sup>1</sup>&emsp;
    <a href="https://scholar.google.com.sg/citations?user=fMXnSGMAAAAJ&hl=en/" target='_blank'>Qingyi Tao</a><sup>2</sup>&emsp;
</div>
<div>
    <sup>1</sup>S-Lab, Nanyang Technological University&emsp; 
    <sup>2</sup>SenseTime Research, Singapore&emsp; 
    <br>
    <sup>†</sup>Project lead
</div>


<div>
    <h4 align="center">
        <a href="https://pq-yang.github.io/projects/MatAnyone2/" target='_blank'>
        <img src="https://img.shields.io/badge/😈-Project%20Page-blue">
        </a>
        <a href="https://arxiv.org/abs/2512.11782" target='_blank'>
        <img src="https://img.shields.io/badge/arXiv-2501.14677-b31b1b.svg">
        </a>
        <a href="https://www.youtube.com/watch?v=tyi8CNyjOhc&lc=Ugw1OS7z5QbW29RZCFZ4AaABAg" target='_blank'>
        <img src="https://img.shields.io/badge/Demo%20Video-%23FF0000.svg?logo=YouTube&logoColor=white">
        </a>
        <a href="https://huggingface.co/spaces/PeiqingYang/MatAnyone" target='_blank'>
        <img src="https://img.shields.io/badge/Demo-%F0%9F%A4%97%20Hugging%20Face-blue">
        </a>
        <a href="https://colab.research.google.com/drive/1NYW_CUDf7jnzxir7tOOlY7wRRalVOifD?usp=sharing" target='_blank'>
        <img src="https://colab.research.google.com/assets/colab-badge.svg">
        </a>
    </h4>
</div>

<strong>MatAnyone 2 is a practical human video matting framework that preserves fine details by avoiding segmentation-like boundaries, while also shows enhanced robustness under challenging real-world conditions.</strong>

<div style="width: 100%; text-align: center; margin:auto;">
    <img style="width:100%" src="https://github.com/pq-yang/MatAnyone2/blob/main/assets/teaser.jpg?raw=true">
</div>

🎥 For more visual results, go checkout our <a href="https://pq-yang.github.io/projects/MatAnyone2/" target="_blank">project page</a>

</div>

---

## How to use
you can run the following commands to get started and start working with the model
```
pip install -qqU git+https://github.com/pq-yang/MatAnyone2.git
```

```python
from matanyone2 import MatAnyone2, InferenceCore
model = MatAnyone2.from_pretrained("PeiqingYang/MatAnyone2")
processor = InferenceCore(model,device="cuda:0")
# inference
processor.process_video(input_path="inputs/video/test-sample2.mp4",
                        mask_path="inputs/mask/test-sample2.png",
                        output_path="results")
```