Search papers, labs, and topics across Lattice.
1
0
3
Force your VLMs to *show their work*: Saliency-R1 aligns model attention with human-annotated visual cues, boosting faithfulness and interpretability without extra compute.