Search papers, labs, and topics across Lattice.
University of Electronic Science and Technology of China
1
0
3
Video-LLMs hallucinate because they fixate on a single "anchor frame," but a simple decoder-side attention fix can dramatically improve grounding without retraining.
