Search papers, labs, and topics across Lattice.
2
0
6
Stop treating videos as monolithic narratives: factorizing video descriptions into streams grounded by identity and time unlocks better video understanding and generation, even closing the gap between small and large MLLMs.
