Search papers, labs, and topics across Lattice.
3
0
6
2
MLLM-based retrieval can dramatically enhance accuracy in visual question answering by focusing on semantic alignment rather than mere visual similarity.
GeoAnchor achieves superior 3D spatial reasoning by integrating multiple latent representations, outperforming existing methods that struggle with complex geometric scenarios.
Current models struggle with hybrid interface tasks, achieving only a 41.2% success rate, underscoring a critical gap in CUA evaluation.
