Self-Supervised Learning

Preprint 2025 - Scaling vision SSL to 7B parameters and 1.7B images, achieving unprecedented patch feature quality.

Oriane Siméoni, Huy v. Vo, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Michaël Ramamonjisoa, Francisco Massa, Daniel Haziza, Luca Wehrstedt, Jianyuan Wang, Timothée Darcet, Théo Moutakanni, Leonel Sentana, Claire Roberts, Andrea Vedaldi, Jamie Tolan, John Brandt, Camille Couprie, Julien Mairal, Hervé Jégou, Patrick Labatut, Piotr Bojanowski

Cluster and Predict Latents Patches for Improved Masked Image Modeling

TMLR 2025 - Stable training of dense image representations using a clustering loss on ViT patch tokens.

Timothée Darcet, Federico Baldassarre, Maxime Oquab, Julien Mairal, Piotr Bojanowski

Cluster and Predict Latents Patches for Improved Masked Image Modeling

Variable Rate Allocation for Vector-Quantized Autoencoders

ICASSP 2023 - Control the bit rate allocation of a a PQ-VAE using semantic information from DINO, in collaboration with Meta AI in Paris.

Federico Baldassarre, Alaaeldin El-Nouby, Hervé Jégou

Variable Rate Allocation for Vector-Quantized Autoencoders

Towards Self-Supervised Learning of Global and Object-Centric Representations

ICLR Workshop 2022 - How to learn object-centric representations using contrastive learning on multi-object datasets?

Federico Baldassarre, Hossein Azizpour