Maxime Oquab

Timeline

1. 2013

   A Microsoft Research internship

   In 2013, Oquab completes a three-month internship at Microsoft Research in New York City.

2. 2014

   Transferring CNN features

   In 2014, he coauthors a CVPR paper that transfers CNN image representations learned on large annotated datasets to object and action recognition with limited training data, and reports results on object and action localization.

3. 2014–2018

   A PhD in visual recognition

   He starts a PhD in 2014 on annotating dynamic scenes with convolutional neural networks, advised by L. Bottou (MSR), I. Laptev, and J. Sivic. He earns the PhD in 2018 with a thesis on using convolutional neural networks for less-supervised visual recognition.

4. 2016

   Classifier tests for GANs

   In 2016, he and David Lopez-Paz coauthor a paper that uses classifier two-sample tests to evaluate generative models such as GANs and applies GANs with the tests to causal discovery.

5. 2023

   DINOv2 visual features

   The 2023 DINOv2 paper shows that self-supervised pretraining on enough curated images can produce visual features that work across image distributions and tasks without fine-tuning.

6. 2024

   An ICLR Outstanding Paper

   He shares ICLR’s 2024 Outstanding Paper award with Timothée Darcet, Julien Mairal, and Piotr Bojanowski for Vision Transformers Need Registers. The paper identifies high-norm tokens in low-information background areas of vision-transformer feature maps and adds register tokens to address them.

7. August 2025

   DINOv3 vision models

   In August 2025, Oquab coauthors DINOv3, a self-supervised vision model that scales data and model size and produces dense features across vision tasks.

8. As of October 2026

   Research scientist at Meta AI

   As of October 2026, he is a research scientist at Meta AI. In August 2026, he coauthors RayDINO, a self-supervised visual encoder for chest X-rays, and provides intellectual advice on model development and supervises the study.