Meta Researcher Proposes CAPI, a New Self-Supervised Learning Method for Masked Image Modeling

iScienceLuvr · x · 2026-07-28

Meta researcher Tim Darcet, known for DINOv2, has proposed a new self-supervised learning method called CAPI (Cluster and Predict Latent Patches) to improve masked image modeling. The paper will be presented at MedARC Journal Club on July 30. MedARC believes CAPI could advance medical imaging foundation models beyond DINO and MAE paradigms.

Original post →

More from Research

Research channel →