Ruifeng Li

Object-conditioned cues can dramatically enhance 3D reconstruction quality, especially under tight frame budgets, outperforming traditional geometry-focused strategies.

Forget expensive data collection: Seed2Scale leverages a small-model/large-model synergy to self-generate high-quality embodied AI training data, starting from just four seed demonstrations.

LiDAR's metric accuracy can be effectively transferred to multi-view stereo depth estimation, yielding state-of-the-art performance in autonomous driving scenarios with improved temporal stability and cross-domain transfer.