PeopleofSuperintelligence

Researcher

Piotr Bojanowski

Piotr Bojanowski is an AI researcher whose work spans fastText and self-supervised visual learning. From 2012 to 2016, he was a computer science PhD student in Inria Paris's WILLOW team; he later joined Facebook AI Research, coauthored fastText papers and wrote about AI for visual storytelling in virtual reality. He coauthored DeepCluster, SwAV, DINO and DINOv2, then became the final author of DINOv3 and coauthored V-JEPA 2 for video understanding and robot planning. Vision Transformers Need Registers won one of five ICLR 2024 Outstanding Paper awards and introduced register tokens for vision transformers. As of October 2026, he is Research Director at Meta and Research Scientist Director at Meta FAIR in Paris; he coauthored papers on global canopy-height mapping and chest X-rays.

Now at
CNRS
Works
88
Citations
36,176
h-index
38
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2012–2016

    PhD in Inria's WILLOW team

    From 2012 to 2016, he is a computer science PhD student in Inria Paris's WILLOW team, supervised by Jean Ponce, Cordelia Schmid, Ivan Laptev and Josef Sivic. In 2015, he is first author of Weakly-Supervised Alignment of Video With Text, which automatically provides a time stamp for every sentence in a video's description.

  2. 2016–2017

    fastText at Facebook AI Research

    From 2016 to 2017, he is a postdoctoral researcher at Facebook AI Research. In 2016, he is first author of Enriching Word Vectors with Subword Information with Edouard Grave, Armand Joulin and Tomas Mikolov; the paper represents each word as a bag of character n-grams, and he coauthors Bag of Tricks for Efficient Text Classification with Armand Joulin, Edouard Grave and Tomas Mikolov, whose fastText classifier is often on par with deep learning classifiers in accuracy and many orders of magnitude faster. FAIR open-sources fastText in 2016 for text representation and classification.

  3. 2017

    AI for visual storytelling in VR

    In 2017, he coauthors the Engineering at Meta article Using AI for new visual storytelling techniques in VR.

  4. 2018–2021

    DeepCluster, SwAV and DINO

    In 2018, he coauthors DeepCluster with Mathilde Caron, Armand Joulin and Matthijs Douze; it uses k-means to group network features and treats the assignments as supervision. In 2020, he coauthors SwAV, which predicts the cluster assignment of one image view from the representation of another view. In 2021, he coauthors DINO, whose self-supervised Vision Transformer features reach 78.3% top-1 on ImageNet as k-NN classifiers with a small ViT.

  5. 2023

    DINOv2 visual features

    In 2023, he is the final author of DINOv2 from Meta AI Research's FAIR lab. The work trains a ViT model with 1B parameters and distills it into a series of smaller models that surpass OpenCLIP on most benchmarks at image and pixel levels.

  6. 2024

    Vision Transformers Need Registers

    In 2024, Vision Transformers Need Registers, coauthored with Timothée Darcet, Maxime Oquab and Julien Mairal, wins one of five ICLR Outstanding Paper awards. It identifies high-norm tokens in low-information background regions of vision-transformer feature maps and addresses them with additional register tokens.

  7. 2025

    DINOv3 and V-JEPA 2

    In 2025, he is the final author of DINOv3, whose Gram anchoring method addresses dense feature maps degrading during long training schedules. He coauthors V-JEPA 2, trained on over 1 million hours of internet video; its post-trained V-JEPA 2-AC plans robot pick-and-place tasks on Franka arms.

  8. As of October 2026

    Research Director at Meta

    As of October 2026, he is Research Director at Meta and Research Scientist Director at Meta FAIR in Paris. In 2026, he coauthors CHMv2, a Scientific Data paper that uses a DINOv3 depth model to build a global meter-resolution canopy-height map. He also coauthors a 2026 Nature Communications paper that introduces RayDINO, a self-supervised visual encoder for chest X-rays.

Career

  1. 2022–2025Institut polytechnique de Grenoble
  2. 2022–2025Inria
  3. 2016–2017FAIR (Meta)Postdoctoral researcher
  4. 2016–2025Meta
  5. 2013–2016Centre Inria de Paris
  6. 2013–2026CNRS
  7. 2013–2017Université Paris Sciences et Lettres
  8. 2013–2017Institut des Sciences de l'Information et de leurs Interactions

Education

  1. 2016InriaPhD student

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Enriching Word Vectors with Subword Information2017 · Transactions of the Association for Computational Linguistics · 9,643 citationsWith Tomas Mikolov
  2. Emerging Properties in Self-Supervised Vision Transformers2021 · IEEE/CVF International Conference on Computer Vision (ICCV) · 5,874 citationsWith Hervé Jégou and Julien Mairal
  3. Bag of Tricks for Efficient Text Classification2017 · Conference of the European Chapter of the Association for Computational Linguistics · 4,538 citationsWith Tomas Mikolov
  4. Deep Clustering for Unsupervised Learning of Visual Features2018 · Lecture notes in computer science · 2,671 citationsWith Matthijs Douze
  5. Unsupervised Learning of Visual Features by Contrasting Cluster Assignments2020 · 2,499 citationsWith Julien Mairal and Priya Goyal

From OpenAlex, as of 21 October 2026.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Piotr Bojanowski. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  2. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes CNRS. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  3. Universities keep AI researchers for a median of 6 years, compared with 2 years at frontier labsIncludes CNRS. Across 15,811 dated work stays (6,529 people), median role tenure runs from 6 years at universities to 2 years at frontier labs. We compare all five organization types, starting decades from the 1970s to the 2020s, and the organizations that keep people longest.
  4. People shaping AI work at a median of 3 organizations over a career: 3 for university starters and institute and lab starters, versus 2 for company startersIncludes CNRS. Across 6,853 people shaping AI with a dated work role, we estimate a median of 3 parent organizations over a career (mean 3.6), comparing role groups, starting sectors and starting decades from the 1950s to 2020–2026.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Piotr Bojanowski yet. Send us a free-licensed photo.

  1. Piotr Bojanowski | alphaXiv alphaxiv.org. Checked 21 October 2026.
  2. [1505.06027] Weakly-Supervised Alignment of Video With Text arxiv.org. Checked 21 October 2026.
  3. [1607.04606] Enriching Word Vectors with Subword Information arxiv.org. Checked 21 October 2026.
  4. [1607.01759] Bag of Tricks for Efficient Text Classification arxiv.org. Checked 21 October 2026.
  5. FAIR open-sources fastText - Engineering at Meta engineering.fb.com. Checked 21 October 2026.
  6. Using AI for new visual storytelling techniques in VR - Engineering at Meta engineering.fb.com. Checked 21 October 2026.
  7. [1807.05520] Deep Clustering for Unsupervised Learning of Visual Features arxiv.org. Checked 21 October 2026.
  8. [2006.09882] Unsupervised Learning of Visual Features by Contrasting Cluster Assignments arxiv.org. Checked 21 October 2026.
  9. [2104.14294] Emerging Properties in Self-Supervised Vision Transformers arxiv.org. Checked 21 October 2026.
  10. [2304.07193] DINOv2: Learning Robust Visual Features without Supervision arxiv.org. Checked 21 October 2026.
  11. GitHub - facebookresearch/dinov2: PyTorch code and models for the DINOv2 self-supervised learning method. github.com. Checked 21 October 2026.
  12. ICLR 2024 Outstanding Paper Awards – ICLR Blog blog.iclr.cc. Checked 21 October 2026.
  13. ICLR Oral Vision Transformers Need Registers iclr.cc. Checked 21 October 2026.
  14. [2309.16588] Vision Transformers Need Registers arxiv.org. Checked 21 October 2026.
  15. Vision Transformers Need Registers arxiv.org. Checked 21 October 2026.
  16. [2508.10104] DINOv3 arxiv.org. Checked 21 October 2026.
  17. DINOv3 arxiv.org. Checked 21 October 2026.
  18. [2506.09985] V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning arxiv.org. Checked 21 October 2026.
  19. V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning arxiv.org. Checked 21 October 2026.
  20. V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning ar5iv.labs.arxiv.org. Checked 21 October 2026.
  21. AI at Scale for Clinical Impact (ASCI): Cancer Pathology Foundation Models asci.artificialintelligencepathology.org. Checked 21 October 2026.
  22. SEA – ICCV 2025 Workshop sustain-eo-ai.github.io. Checked 21 October 2026.
  23. CHMv2: Improvements in Global Canopy Height Mapping using DINOv3 | Scientific Data nature.com. Checked 21 October 2026.
  24. [2603.06382] CHMv2: Improvements in Global Canopy Height Mapping using DINOv3 arxiv.org. Checked 21 October 2026.
  25. Advancing human-centric AI for robust X-ray analysis through holistic self-supervised learning - PMC pmc.ncbi.nlm.nih.gov. Checked 21 October 2026.
  26. Advancing human-centric AI for robust X-ray analysis through holistic self-supervised learning | Nature Communications nature.com. Checked 21 October 2026.