Home

Awesome

Real-Time High-Resolution Background Matting

Teaser

Official repository for the paper Real-Time High-Resolution Background Matting. Our model requires capturing an additional background image and produces state-of-the-art matting results at 4K 30fps and HD 60fps on an Nvidia RTX 2080 TI GPU.

Disclaimer: The video conversion script in this repo is not meant be real-time. Our research's main contribution is the neural architecture for high resolution refinement and the new matting datasets. The inference_speed_test.py script allows you to measure the tensor throughput of our model, which should achieve real-time. The inference_video.py script allows you to test your video on our model, but the video encoding and decoding is done without hardware acceleration and parallization. For production use, you are expected to do additional engineering for hardware encoding/decoding and loading frames to GPU in parallel. For more architecture detail, please refer to our paper.

 

New Paper is Out!

Check out Robust Video Matting! Our new method does not require pre-captured backgrounds, and can inference at even faster speed!

 

Overview

 

Updates

 

Download

Model / Weights

Video / Image Examples

Datasets

 

Demo

Scripts

We provide several scripts in this repo for you to experiment with our model. More detailed instructions are included in the files.

Notebooks

Additionally, you can try our notebooks in Google Colab for performing matting on images and videos.

Virtual Camera

We provide a demo application that pipes webcam video through our model and outputs to a virtual camera. The script only works on Linux system and can be used in Zoom meetings. For more information, checkout:

 

Usage / Documentation

You can run our model using PyTorch, TorchScript, TensorFlow, and ONNX. For detail about using our model, please check out the Usage / Documentation page.

 

Training

Configure data_path.pth to point to your dataset. The original paper uses train_base.pth to train only the base model till convergence then use train_refine.pth to train the entire network end-to-end. More details are specified in the paper.

 

Project members

<sup>* Equal contribution.</sup>

 

License

This work is licensed under the MIT License. If you use our work in your project, we would love you to include an acknowledgement and fill out our survey.

Community Projects

Projects developed by third-party developers.