Search papers, labs, and topics across Lattice.
1
0
3
Current VLM spatial reasoning benchmarks are misleading, as they often penalize models for "incorrect" answers that are actually correct given the limited visual information the models receive.
