PeopleofSuperintelligence

Researcher

Heng Wang

Heng Wang is a computer vision researcher whose work on video understanding runs from hand-crafted motion features to world models. After a PhD in computer vision at the Chinese Academy of Sciences, he presented the dense trajectory feature in a CVPR paper in 2011 and improved it in 2013 as a postdoc in the LEAR team at INRIA, working with Cordelia Schmid. In November 2014, he joined Amazon in Seattle as an early member of the Amazon Go team; at Facebook AI Research, he coauthored the R(2+1)D network in 2017 and the TimeSformer video transformer in 2021. He received the Helmholtz Prize at ICCV 2023 and worked at Nvidia on the Cosmos world models released in January 2025. Today he is a research scientist at Tencent, focused on world models.

Now at
Tencent

Timeline

  1. 2011

    Dense trajectories for video

    He earns a B.S. in electrical engineering at Harbin Institute of Technology and a PhD in computer vision from the Chinese Academy of Sciences. In 2011, he presents the dense trajectory feature in a CVPR paper.

  2. 2013

    Improved trajectories at INRIA

    As a postdoc in the LEAR team at INRIA, working with Cordelia Schmid, he coauthors “Action Recognition with Improved Trajectories” with Schmid at ICCV 2013. The method estimates camera motion from SURF matches and dense optical flow with RANSAC, then removes it to improve action recognition on Hollywood2, HMDB51, Olympic Sports and UCF50.

  3. November 2014

    Amazon Go in Seattle

    He becomes a research scientist at Amazon in Seattle in November 2014. He is an early member of the Amazon Go team, which builds a computer vision system to replace human cashiers in retail.

  4. 2017

    The R(2+1)D network

    In 2017, he coauthors “A Closer Look at Spatiotemporal Convolutions for Action Recognition” with Du Tran, Lorenzo Torresani, Jamie Ray, Yann LeCun and Manohar Paluri. The paper factorizes 3D convolutions into separate spatial and temporal parts, leading to the R(2+1)D block. He works as a research scientist at Facebook AI Research.

  5. 2021

    TimeSformer and PyTorchVideo

    With Gedas Bertasius and Lorenzo Torresani, he coauthors TimeSformer in 2021, a convolution-free video classifier built on self-attention over space and time and accepted to ICML 2021. PyTorchVideo, an open-source PyTorch library for video understanding that he coauthors, appears in 2021.

  6. 2023

    A Helmholtz Prize

    He receives the Helmholtz Prize, a test-of-time award, at ICCV 2023. In September 2023, he coauthors a paper on his team's winning DataComp challenge entry, a three-stage data-filtering strategy for pre-training foundation models.

  7. January 2025

    Cosmos world models at Nvidia

    As a research scientist at Nvidia, he works on the Cosmos world models, released at CES in 2025. In January 2025, he coauthors the Cosmos World Foundation Model Platform for Physical AI paper, which presents a video curation pipeline, pre-trained world foundation models and video tokenizers.

  8. As of October 2026

    Research scientist at Tencent

    He is a research scientist at Tencent, with interests from visual understanding to generation and a recent focus on world models. Before Tencent, he works as a research lead at TikTok/ByteDance building Seed foundation models.

Career

  1. NowTencentResearch Scientist
  2. From 2025NVIDIAResearch Scientist
  3. From 2017FAIR (Meta)Research Scientist
  4. From 2014AmazonResearch Scientist
  5. From 2013InriaPostdoctoral Researcher
  6. ByteDanceResearch Lead

Education

  1. Harbin Institute of TechnologyB.S.
  2. Chinese Academy of SciencesPhD

From the biography.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Heng Wang. 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 Amazon. 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 Tencent. 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. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Tencent. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.

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 Heng Wang yet. Send us a free-licensed photo.

  1. About Me - Heng Wang’s Homepage hengcv.github.io. Checked 21 October 2026.
  2. THOTH - Heng Wang lear.inrialpes.fr. Checked 21 October 2026.
  3. Action Recognition with Improved Trajectories - ICCV 2013 Open Access Repository openaccess.thecvf.com. Checked 21 October 2026.
  4. [1711.11248] A Closer Look at Spatiotemporal Convolutions for Action Recognition arxiv.org. Checked 21 October 2026.
  5. [2102.05095] Is Space-Time Attention All You Need for Video Understanding? arxiv.org. Checked 21 October 2026.
  6. [2111.09887] PyTorchVideo: A Deep Learning Library for Video Understanding arxiv.org. Checked 21 October 2026.
  7. [2309.15954] The Devil is in the Details: A Deep Dive into the Rabbit Hole of Data Filtering arxiv.org. Checked 21 October 2026.
  8. [2501.03575] Cosmos World Foundation Model Platform for Physical AI arxiv.org. Checked 21 October 2026.