Human-centric Video Matting Challenge

Powered by open dataset


The outbreak of the coronavirus this past year has accelerated the adoption of video conferencing software. More and more academia, companies and governments are using video conference software for various purposes.Video conference application is enabling a new era for computer vision on research and innovation. is working on an open dataset specially designed for video conferencing, to support innovative research, such as setting virtual backgrounds, instance segmentation, sematic segmentation and visual relationship detection.

We are very glad to announce our release of the open dataset for Human-centric video matting ( ), in collaborating with Future Video Conferencing (FVC) at CVPR 2021.


  • April 9, 2021 – Start Date
  • June 4, 2021 – End Date
Dataset Description

We have collected data from various video conferencing APPs on both desktops and mobile phones. Videos are taken indoors (office,home,cafe ect.) and outdoors (park,street,car,train,etc.).

Each video has been trimmed to ~15s using our interactive video segmentation tools.

Licenses is providing this open dataset under the CC BY-NC 4.0 license.


Yunchao Wei

Humphrey Shi

Zhangyang Wang

Jiaying Liu

Vicky Kalogeiton

Jiashi FENG

Marta Mrak


UT Austin

Robotic X Lab and AI Lab

University of Washington


Google Research

Australian National University

University of Minnesota


Please feel free to contact if you have any questions.
Scan the QR code to join the groups.

 Wechat group

 WhatsApp group

Other Competition