Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
3
0
4
Ditch the blurry latent space: MLLMs can guide diffusion models to demorph faces directly in RGB, outperforming latent-space methods by 30-40%.
Multi-image reasoning in Vision-Language Models can be dramatically improved by leveraging a novel preference optimization framework that emphasizes global visual search over localized tasks.
Face embeddings from different DNNs, even those trained on different data and with different architectures, can be surprisingly well-aligned with simple linear transformations, unlocking cross-model face recognition.