Search papers, labs, and topics across Lattice.
3
0
4
Self-distillation using SAM3 as a teacher and innovative multi-scale augmentation led to a surprising 69.73% mIoU, showcasing the power of leveraging existing models for enhanced performance.
Giving VLMs access to basic image manipulation tools and a strategic routing system dramatically improves their ability to "see through" visual illusions, even generalizing to unseen illusion types.
MLLMs can generate surprisingly effective synthetic training data for defect classification, boosting performance by 20% even with very limited real data.
