Search papers, labs, and topics across Lattice.
3
0
5
9
Over 1,500 submissions revealed stark differences in model performance across diverse domains, highlighting the challenges of generalizing egocentric video understanding.
Latent objects can revolutionize how Video-LLMs manage memory, achieving a 10-point performance boost while slashing memory usage by 50%.
By explicitly formulating and verifying hypotheses *before* retrieving information, VideoHV-Agent side-steps the semantic drift and correlation-driven errors that plague traditional long-video QA systems.
