Search papers, labs, and topics across Lattice.
4
0
5
4
ProMSA achieves superior accuracy in KB-VQA by dynamically selecting retrieval strategies, outperforming traditional fixed pipelines.
ViQ achieves a groundbreaking balance between semantic richness and detail in visual representations, enabling efficient multimodal training without sacrificing quality.
LiveStarPro achieves a 28.9% boost in semantic correctness for long-horizon video understanding while maintaining real-time responsiveness.
MemDreamer narrows the performance gap with human experts in long video understanding to just 3.7 points while processing only 2% of the full context.
