Search papers, labs, and topics across Lattice.
4
0
7
17
Synthesizing realistic human-object interactions is now possible by directly coupling generative human motion with explicit physics simulation, opening the door to more believable virtual environments.
Diffusion models get a free lunch: a new sampler that cuts FID scores by up to 20% just by injecting the right kind of noise at each step.
Achieve competitive novel-view synthesis with a 4MB footprint and 78ms inference by learning a compact global scene representation *before* decoding 3D geometry.
Forget training: SemanticMoments achieves state-of-the-art motion-based video retrieval by simply computing temporal statistics over features from pre-trained semantic models.
