Search papers, labs, and topics across Lattice.
Affiliation:
2
0
5
2
RefineCut's innovative verifier-grounded approach elevates video-editing performance to new heights, achieving a score of 0.924 while eliminating the need for teacher calls at inference.
VISA's self-evolving framework not only enhances multimodal instruction synthesis but also adapts in real-time to improve training data quality and model performance.
