Our group, the Precognition Group (智能感知与预测研究组), is interested in building human-level AI systems. We focus on machine perception of human activities and the world, jointly with machine prediction of unseen future states. We believe in order to enable machines to understand and behave like humans, we should build world models with common sense knowledge that could aid intelligent agents in making action decisions. As a first step, sequential data (like that of videos) perception and prediction are important research topics to tackle.
- 01/2023 I am co-organizing the The 5th workshop on Precognition: Seeing through the Future @CVPR 2023.
- 10/2022 Two papers accepted at NeurIPS 2022. [Multi-Action (Spotlight paper, 3.7% acceptance rate, 384/10411)] [Video Retrieval]
- 06/2022 Achieved second-place out of 150 teams on the public leaderboard of the Naturalist Driver Action Recognition Task - AI City Challenge @ CVPR 2022. [CVPRW Paper] [Presentation] [Code and Model]
- 10/2021 Published a research talk at TechBeat.net on Pedestrian Trajectory Prediction. [将门TechBeat] [B站]
- 08/2021 1 paper accepted by ICCV 2021.
- 08/2021 Our VERA system helps another major Washington Post news report. [link]
- 04/2021 Featured in a front-page news report (04/15) by Washington Post using crowding counting technologies. [video] [知乎]
- 01/2021 Invited presentation at ICPR'20 pattern forecasting workshop. [link]
- 09/2020 We won the Automated Streams Analysis for Public Safety Challenge with a $30k prize.
- 08/2020 Our paper has been accepted by WACV 2021 (one strong-accept) and reported by CMU news:
- 08/2020 Analyzed videos for journalist from the Washington Post on a major news.
- 07/2020 SimAug paper accepted by ECCV 2020.
- 06/2020 Multiverse (CVPR 2020) code and dataset are released! [blog] [知乎] [code]
09/2019 Our Shooter Localization System won Best Demo award at CBMI2019. [Project Site]
Press Coverage: , , , , ,
- 06/2019 Presented Future Prediction paper at CVPR 2019. It was reported by the media and it received 30k+ views in a week. [Tweets]
- 04/2019 Our CMU team's (INF & MUDSML) system achieved the best performance on the activity detection challenge (Cached) in surveillance videos hosted by NIST & IARPA. We have released our code and model for Object Detection & Tracking here.
- 12/2018 MemexQA paper accepted by TPAMI 2019.
- 06/2018 Presented MemexQA paper at CVPR 2018. [Spotlight Talk]
- [03/2017] Two papers accepted by ICASSP 2017.
- [02/2017] Two demo papers accepted by AAAI 2017.
- 11/2016 Best performer in the NIST TRECVID 2016 Ad-hoc Video Search Challenge (no annotation track).
- [02/2016] One oral paper accepted by IJCAI 2016.
- Pedestrian trajectory prediction. [Project Page]