Search papers, labs, and topics across Lattice.
1
0
2
DINOv3 achieves 83% accuracy in region-level facial correspondence, outperforming random baselines and revealing the power of frozen models in zero-shot settings.
