PeopleofSuperintelligence

Researcher

Haoqi Fan

Haoqi Fan is a computer vision researcher whose work spans video recognition, unsupervised representation learning and unified multimodal models. After studying in Carnegie Mellon's Robotics Institute, he worked at Facebook AI Research from 2017 to 2023, where he co-authored SlowFast networks in 2019, Momentum Contrast in 2020 (a CVPR best paper nominee) and Multiscale Vision Transformers in 2021. He joined ByteDance Seed in 2023 and co-authored the open-source BAGEL model, released in May 2025. Since 2026, he heads a world model research group at Seed.

Now at
ByteDance
Badges

Timeline

  1. 2015–2017

    Robotics Institute at Carnegie Mellon

    He is an MSc student at Carnegie Mellon University from 2015 to 2016 and graduates from its Robotics Institute. From 2016 to 2017, he is a research engineer in the Robotics Institute.

  2. 2017–2023

    Researcher at Facebook AI Research

    He is a computer vision researcher at Facebook AI Research (FAIR) from 2017 to 2023, working on deep learning, video understanding and multimodal foundational models.

  3. 2019

    SlowFast networks and PySlowFast

    He co-authors SlowFast networks for video recognition with Christoph Feichtenhofer, Jitendra Malik and Kaiming He, pairing a Slow pathway at low frame rate for spatial semantics with a Fast pathway at high frame rate for motion. The paper is an oral at ICCV 2019. He wins first place in the AVA video activity detection challenge at the ICCV 2019 International Challenge on Activity Recognition in June 2019 and releases the PySlowFast codebase for video understanding research in November 2019.

  4. 2020

    Momentum Contrast and a CVPR nomination

    He co-authors Momentum Contrast (MoCo) with Kaiming He, Yuxin Wu, Saining Xie and Ross Girshick; it builds a dynamic dictionary with a queue and a moving-averaged encoder for unsupervised visual representation learning. The paper is an oral at CVPR 2020 and receives a Best Paper Nomination.

  5. 2021

    Multiscale Vision Transformers and PyTorchVideo

    He is the first author of Multiscale Vision Transformers (MViT), which connects multiscale feature hierarchies with transformer models for video and image recognition and appears at ICCV 2021. In April 2021, he releases PyTorchVideo, a deep learning library for video understanding research.

  6. 2023

    Joining ByteDance Seed

    He moves from Facebook AI Research to ByteDance Seed in 2023 and works there as a researcher.

  7. May 2025

    BAGEL unified multimodal model

    He co-authors BAGEL, an open-source decoder-only model pretrained on trillions of tokens of interleaved text, image, video and web data for multimodal understanding and generation. Released in May 2025, it quickly becomes the number one trending model on Hugging Face.

  8. Since 2026

    World model group at Seed

    After the Spring Festival in 2026, ByteDance Seed sets up a new world model research group led by him, reporting to Zhou Chang, Seed's head of multimodal and world model research. The group takes a 3D simulation route aimed at entertainment and games.

Career

  1. Since 2023ByteDanceResearcher, Head of World Model Research Group
  2. 2017–2023FAIR (Meta)Computer Vision Researcher
  3. 2016–2017Carnegie Mellon UniversityResearch Engineer

Education

  1. 2016Carnegie Mellon UniversityMSc

From the biography.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Haoqi Fan. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes ByteDance. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  3. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes ByteDance. 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.
  4. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes ByteDance. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.

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 Haoqi Fan yet. Send us a free-licensed photo.

  1. Haoqi Fan haoqifan.github.io. Checked 21 October 2026.
  2. Haoqi Fan | alphaXiv alphaxiv.org. Checked 21 October 2026.
  3. [1812.03982] SlowFast Networks for Video Recognition arxiv.org. Checked 21 October 2026.
  4. [1911.05722] Momentum Contrast for Unsupervised Visual Representation Learning arxiv.org. Checked 21 October 2026.
  5. [2104.11227] Multiscale Vision Transformers arxiv.org. Checked 21 October 2026.
  6. GitHub - facebookresearch/pytorchvideo: A deep learning library for video understanding research. github.com. Checked 21 October 2026.
  7. [2505.14683] Emerging Properties in Unified Multimodal Pretraining arxiv.org. Checked 21 October 2026.
  8. 36Kr Exclusive: Four Key Propositions of ByteDance AI in 2026 eu.36kr.com. Checked 21 October 2026.
  9. ByteDance sets four AI priorities for 2026 kr-asia.com. Checked 21 October 2026.