Search papers, labs, and topics across Lattice.
B LLM consistently underperforming compared to the
2
0
4
Forget slow generative models: WikiCLIP delivers a 16% accuracy boost in visual entity recognition with 100x faster inference by cleverly combining CLIP-style contrastive learning with vision-guided knowledge adaptation.
Diffusion models can now generate more realistic and semantically appropriate hand grasps by explicitly modeling affordances and interaction semantics, outperforming prior methods on grasp quality, semantic accuracy, and diversity.
