Wei Mao

Wei Mao 毛伟

I am a Senior Research Scientist at Tencent (Canberra) from Jan 2024 working on 3D AIGC for games. Previously, I was a postdoc at Australian National University (ANU) advised by Prof. Richard Hartley, and Dr. Miaomiao Liu. I also work closely with Dr. Mathieu Salzmann from EPFL. My research interests mainly lie in two folds: human motion understanding including human motion prediction/generation, human-scene interaction; 3D vision including 3D reconstuction of dynamic scenes, neural rendering, and depth estimation. I obtained my Ph.D. from ANU in 2022 supervised by Dr Miaomiao Liu. Prior to that I received a master degree in Computing from ANU in 2018, and a Bachelor of Engineering from East China University of Science and Technology (ECUST) in 2013, respectively.

Email / CV (中文简历) / Google Scholar / Github / Linkedin

News

07/2024: Two papers have been accepted to ECCV2024. Congrates to Rong and Chaoyue.
01/2024: Our paper "Neural SDF Flow for 3D Reconstruction of Dynamic Scenes" has been accepted to ICLR2024.
09/2023: One paper has been accepted to NeurIPS2023, Congrates to Rong.
03/2023: One paper has been accepted to CVPR2023, Congrates to Huiyu.
12/2022: My PhD thesis titled "Human Motion Prediction: From Deterministic to Stochastic" has been published (here).
10/2022: Our paper "Contact-aware Human Motion Forecasting" has been accepted to NeurIPS2022
03/2022: One paper accepted to CVPR2022 (oral)
01/2022: I will work as a Postdoc with Prof. Richard Hartley and Dr. Miaomiao Liu after finishing my PhD !!!
07/2021: One paper has been accepted to ICCV2021 (oral)
03/2020: One paper has been accepted to CVPR2020 (oral). Congrates to Jiayu.

Research

I have broad interests in human related research and 3D vision.

	Neural SDF Flow for 3D Reconstruction of Dynamic Scenes Wei Mao, Miaomiao Liu, Richard Hartley, Mathieu Salzmann ICLR, 2024 openreview / code We propose SDF flow, a new implicit representation for dynamic scenes. We derived a linear relationship between scene flow and SDF flow bridging the scene geometry with its motion.
	DeepSimHO: Stable Pose Estimation for Hand-Object Interaction via Physics Simulation Rong Wang, Wei Mao, Hongdong Li NeurIPS, 2023 openreview / video / code We propose to learn the stability score from physical simulators for predicting hand poses that can firmly grasp the objects.
	VisFusion: Visibility-aware Online 3D Scene Reconstruction from Videos Huiyu Gao, Wei Mao, Miaomiao Liu CVPR, 2023 arXiv / project page / code Estimating visibility aware feature volume with ray-based sparsification for volumetric-based 3D reconstruction.
	Interacting Hand-Object Pose Estimation via Dense Mutual Attention Rong Wang, Wei Mao, Hongdong Li WACV, 2023 arXiv / video / code Mutual attention between hand and object for hand object pose estimation.
	Contact-aware Human Motion Forecasting Wei Mao, Miaomiao Liu, Richard Hartley, Mathieu Salzmann NeurIPS, 2022 (Spotlight) arXiv / video / code Predicting per-joint contact maps for human motion forecasting.
	Weakly-supervised Action Transition Learning for Stochastic Human Motion Prediction Wei Mao, Miaomiao Liu, Mathieu Salzmann CVPR, 2022 (Oral Presentation) arXiv / video / code Generating diverse human motion with a sequence of action labels.
	Generating Smooth Pose Sequences for Diverse Human Motion Prediction Wei Mao, Miaomiao Liu, Mathieu Salzmann ICCV, 2021 (Oral Presentation) arXiv / video / code A model to predict diverse human motions without diverse training samples.
	History Repeats Itself: Human Motion Prediction via Motion Attention Wei Mao, Miaomiao Liu, Mathieu Salzmann ECCV, 2020 arXiv / video / code Capture similar motion pattern with motion attention.
	Cost Volume Pyramid Based Depth Inference for Multi-View Stereo Jiayu Yang, Wei Mao, Jose M. Alvarez, Miaomiao Liu CVPR, 2020 (Oral Presentation) arXiv / video / code A coarse-to-fine approach for multi-view stereo.
	Learning Trajectory Dependencies for Human Motion Prediction Wei Mao, Miaomiao Liu, Mathieu Salzmann Hongdong Li ICCV, 2019 (Oral Presentation) arXiv / video / code Human motion prediction using DCT for temporal encoding and GCN for spatial encoding.

Website template is from Jon Barron