DINO v1 and v2: Self-Supervised Vision Transformers

In this meeting we discussed the self-supervised learning of vision transformers. First half of the presentation was based on the 2021 paper “Emerging Properties in Self-Supervised Vision Transformers”. Second half was about the second version of DINO based on the 2023 paper “DINOv2: Learning Robust Visual Features without Supervision”.

You can find the presentations that were held here and here.