Researchers from Zhejiang University, ByteDance, and The Chinese University of Hong Kong, Shenzhen have developed PointSplat, a novel method for creating more compact 3D human representations. This approach shifts from a view-centric to a human-centric prediction model, aiming to reduce redundancy in 3D Gaussian Splatting (3DGS) by first establishing a rough geometric proxy of the human. By projecting rays onto this proxy and filtering out irrelevant points, the method efficiently fuses multi-view geometric and appearance information to predict Gaussian attributes, thereby addressing the storage and transmission challenges of real-time 3D applications like immersive live streaming. AI
IMPACT This human-centric approach to 3D representation could significantly reduce data requirements for real-time applications like immersive streaming and digital humans.
RANK_REASON Research paper accepted at a major conference (ECCV 2026) detailing a new method for 3D representation. [lever_c_demoted from research: ic=1 ai=1.0]
- 3D Gaussian Splatting
- ByteDance
- ECCV 2026
- I3D 2024
- immersive live streaming
- NeRF
- PointSplat
- SIGGRAPH 2023
- The Chinese University of Hong Kong, Shenzhen
- Zhejiang University
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →