Search papers, labs, and topics across Lattice.
1
0
3
2
MLLMs can get a whopping 8% boost in spatial reasoning accuracy, with 50% less memory, simply by fusing geometric and semantic features in the vision encoder.
