PeopleofSuperintelligence

Researcher

Jie Hu

Also known as 胡杰, Hu Jie

Jie Hu is a computer vision researcher and the first author of Squeeze-and-Excitation Networks (SENet), whose SE block lets convolutional neural networks recalibrate channel-wise feature responses. Working at Momenta in 2017, Hu designed the SE block with Li Shen and Gang Sun, and their WMW team won the image classification task of ImageNet 2017 with a top-5 error of 2.251%. SENet appeared at CVPR 2018 and Gather-Excite at NeurIPS 2018; by 2019, when a TPAMI journal version of SENet was accepted, Hu was with the Institute of Software, Chinese Academy of Sciences. Later papers include Kangaroo, a 2024 video-language model for long-context video input written while Hu worked with Meituan Group, and D-JEPA, a generative model presented at ICLR 2025. They now work at TikTok in Singapore.

Now at
ByteDance and Institute of Software, Chinese Academy of Sciences
Works
54
Citations
49,482
h-index
16
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2017

    Squeeze-and-Excitation at Momenta

    Working at Momenta in 2017 with Li Shen of the University of Oxford and Gang Sun, Hu designs a new architecture building block named Squeeze-and-Excitation (SE). Each SE block embeds information from global receptive fields through a squeeze operation.

  2. 2017

    First place at ImageNet 2017

    An SENet-based entry from the WMW team, made up of Hu, Shen and Sun, wins the image classification task of ImageNet 2017 with a top-5 error of 2.251%. At CVPR 2017 in July 2017, the Beyond ILSVRC workshop, which marks the last of the ImageNet Challenge competitions, features Hu's talk on Squeeze-and-Excitation Networks.

  3. 2018

    SENet at CVPR 2018

    Squeeze-and-Excitation Networks appears at CVPR 2018 with Hu as first author, alongside Li Shen and Gang Sun. Its SE block adaptively recalibrates channel-wise feature responses by modelling interdependencies between channels, and the code and models are released at github.com/hujie-frank/SENet.

  4. 2018

    Gather-Excite at NeurIPS 2018

    Gather-Excite, a pair of lightweight operators that aggregate feature responses over a large spatial extent and redistribute the pooled information to local features, appears at NeurIPS 2018 with Hu, Li Shen, Samuel Albanie, Gang Sun and Andrea Vedaldi as authors. With these operators, a ResNet-50 outperforms its 101-layer counterpart on ImageNet with no additional learnable parameters.

  5. 2019

    SENet in TPAMI

    In May 2019, the journal version of Squeeze-and-Excitation Networks, accepted by TPAMI, adds Samuel Albanie and Enhua Wu as coauthors. The State Key Laboratory of Computer Science at the Institute of Software, Chinese Academy of Sciences, is by then one of Hu's affiliations, along with the University of Chinese Academy of Sciences and Momenta.

  6. 2021–2023

    Binarized and quantized networks

    Elastic-Link, an AAAI 2022 paper posted in December 2021 with Hu as first author, adaptively adds real-valued input features to the convolutional outputs of binarized neural networks. A March 2023 ICASSP paper with Mengze Zeng and Enhua Wu proposes guidelines for post-training quantization of convolutional networks.

  7. 2024–2025

    Kangaroo and D-JEPA

    In August 2024, while working with Meituan Group and the Institute of Software, Hu coauthors Kangaroo, a video-language model built for long-context video input. D-JEPA, written with Dengsheng Chen, Xiaoming Wei and Enhua Wu, brings joint-embedding predictive architectures into generative modelling and appears at ICLR 2025.

  8. As of October 2026

    TikTok in Singapore

    They work at TikTok in Singapore.

Career

  1. Since 2021ByteDance
  2. 2025–2026Meituan
  3. 2025–2026National University of Singapore
  4. 2024–2026Chinese Academy of Sciences
  5. From 2019Institute of Software, Chinese Academy of Sciences
  6. 2019University of Chinese Academy of Sciences
  7. 2018–2025Xiamen University
  8. From 2017Momenta

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

Most-cited works

  1. Squeeze-and-Excitation Networks2018 · IEEE Computer Society Conference on Computer Vision and Pattern Recognition · 35,429 citations
  2. Gather-Excite: Exploiting Feature Context in Convolutional Neural Networks2018 · 574 citationsWith Andrea Vedaldi
  3. Involution: Inverting the Inherence of Convolution for Visual Recognition2021 · IEEE Computer Society Conference on Computer Vision and Pattern Recognition · 447 citations
  4. A Key Volume Mining Deep Framework for Action Recognition2016 · IEEE Computer Society Conference on Computer Vision and Pattern Recognition · 249 citationsWith Yu Qiao
  5. Image-to-image Translation via Hierarchical Style Disentanglement2021 · IEEE Computer Society Conference on Computer Vision and Pattern Recognition · 152 citationsWith Rongrong Ji

From OpenAlex, as of 21 October 2026.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes ByteDance. 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. Universities keep AI researchers for a median of 6 years, compared with 2 years at frontier labsIncludes Chinese Academy of Sciences. 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.
  3. 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.
  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 Chinese Academy of Sciences. 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 Jie Hu yet. Send us a free-licensed photo.

  1. ILSVRC2017 results image-net.org. Checked 20 October 2026.
  2. Squeeze-and-Excitation Networks (CVPR 2018 Open Access Repository) openaccess.thecvf.com. Checked 20 October 2026.
  3. Beyond ILSVRC workshop 2017 image-net.org. Checked 20 October 2026.
  4. Gather-Excite: Exploiting Feature Context in Convolutional Neural Networks proceedings.neurips.cc. Checked 20 October 2026.
  5. [1709.01507] Squeeze-and-Excitation Networks arxiv.org. Checked 20 October 2026.
  6. [1709.01507] Squeeze-and-Excitation Networks (ar5iv) ar5iv.labs.arxiv.org. Checked 20 October 2026.
  7. [2112.10149] Elastic-Link for Binarized Neural Network arxiv.org. Checked 20 October 2026.
  8. [2303.07080] Bag of Tricks with Quantized Convolutional Neural Networks for image classification arxiv.org. Checked 20 October 2026.
  9. Kangaroo: A Powerful Video-Language Model Supporting Long-context Video Input arxiv.org. Checked 20 October 2026.
  10. Denoising with a Joint-Embedding Predictive Architecture proceedings.iclr.cc. Checked 20 October 2026.
  11. hujie-frank (Jie Hu) · GitHub github.com. Checked 20 October 2026.