speridlabs / iris-3b

Generative priors are a promising foundation for downstream vision tasks. In this project we explore pixel-space generative models as an alternative to vision foundation models such as DINOv2.