Search papers, labs, and topics across Lattice.
South China University of Technology
2
0
4
Despite achieving 84.8% accuracy, multimodal models struggle with long-horizon reasoning in electrical circuits, revealing critical gaps in their understanding of physical conventions.
Relationship-aware 3D scene understanding can significantly enhance open-vocabulary segmentation, outperforming traditional methods that ignore object relationships.
