04/03/2026
Research Break: Ultra-Fast 2D-to-3D Video Conversion
Our team at The Hong Kong University of Science and Technology (Guangzhou), in collaboration with Kuaishou Kling, introduces StereoPilot. This novel end-to-end model converts 5-second 2D clips to high-fidelity 3D video in just 11 seconds, outperforming SOTA methods.
The work is led by Prof. Ying-Cong Chen (Assistant Professor, AI Thrust, Information Hub), with Guibao Shen, Yihua Du, and Wenhang Ge as co-first authors from the same domain.
Key innovations:
✅ Solves "depth ambiguity" for complex scenes like mirrors.
✅ Unifies 3D formats via a learnable Domain Switcher.
✅ "Diffusion as Feed-Forward" architecture for unprecedented speed.
This significantly lowers the barrier for 3D content creation. Paper & details below.