Juncheng Ma

Juncheng Ma is a first-year Master's student at the School of Electronic and Computer Engineering at Peking University, advised by Prof. Daquan Zhou and Prof. Yonghong Tian. His research focuses on world-action models for embodied intelligence, and he is best known for his work on the HumanScale and Open-HumanWAM projects, which demonstrate that egocentric human video can serve as a scalable pretraining source for embodied foundation models. He has also published work on audio-driven portrait animation and audio-visual semantic segmentation, including papers accepted to ECCV 2024 and ECCV 2026.