Search papers, labs, and topics across Lattice.
Affiliation:
1
0
Task-specific visual token compression, not just saliency, unlocks substantial efficiency gains for VLA models without retraining.
