Search papers, labs, and topics across Lattice.
Zhejiang University Hangzhou
1
0
2
Retaining 79.3% of full performance with just 32 visual tokens reveals that spatial coverage trumps traditional importance metrics in multimodal models.
