Search papers, labs, and topics across Lattice.
University of Illinois Urbana-Champaign
2
0
4
Semantic attribute gradients from frozen MLLMs can boost zero-shot image retrieval performance by up to 6 points, reshaping how visual embeddings are trained.
World Tracing not only reconstructs visible surfaces but also reveals occluded geometry, unlocking advanced capabilities in 3D scene manipulation and synthesis.
