Search papers, labs, and topics across Lattice.
5
0
7
Lift4D achieves unprecedented accuracy in 4D reconstruction of dynamic objects, even in the presence of severe occlusions and complex motions.
Hierarchical sub-goal policies enable robots to adapt to novel tasks with just a few human demonstrations, significantly enhancing their manipulation capabilities.
Unlabeled monocular videos can now be used to train state-of-the-art 3D/4D reconstruction systems, thanks to a factored flow prediction approach that disentangles geometry and pose learning.
Skip task-specific reward engineering: Dex4D learns a single, generalizable dexterous manipulation policy in simulation that transfers zero-shot to real-world tasks by tracking object-centric points.
EditCtrl slashes the compute cost of generative video editing by 10x while *improving* quality, thanks to its focus on local masked tokens and lightweight global context.
