Search papers, labs, and topics across Lattice.
1
0
2
DIVE achieves an impressive 88.9% reduction in visual tokens without sacrificing performance, redefining efficiency in vision-language models.
