Search papers, labs, and topics across Lattice.
Yunnan University,School of Information Science and Engineering,Kunming,China, Netflix
1
1
3
5
Freezing both language and vision encoders, combined with contrastive decoding, unlocks more accurate multimodal chain-of-thought reasoning by forcing models to truly "see" the difference between images.
