Yosuke Shinya

Yosuke Shinya

Biography

Yosuke Shinya is a research engineer in computer vision and deep learning, especially in object detection and image generation. He is a senior researcher at SenseTime Japan. He has received the Award of the Minister of State for Science and Technology Policy (at the 50th Japan Student Science Award), OpenMMLab Contributor of the Year Award 2021, and Papers with Code Top Contributor Award.

進矢陽介:物体検出や画像生成を中心としたコンピュータビジョンの研究開発に従事。 2014年東京大学大学院情報理工学系研究科修士課程修了。 三菱電機株式会社、株式会社デンソーを経て、2024年に株式会社センスタイムジャパンに入社。 主な受賞歴に、第50回日本学生科学賞 科学技術政策担当大臣賞、OpenMMLab Contributor of the Year Award 2021、Papers with Code Top Contributor Award。

Interests
  • Computer Vision
  • Deep Learning
  • Computer Graphics
  • Augmented Reality
Education
  • Master of Information Science and Technology, 2014

    The University of Tokyo

  • Bachelor of Engineering, 2012

    The University of Tokyo

Experience

 
 
 
 
 
Senior Researcher
SenseTime Japan
Jun 2024 – Present Tokyo, Japan
Conducting research and development on computer vision.
 
 
 
 
 
Research Engineer, Project Assistant Manager
DENSO
Dec 2015 – May 2024 Tokyo, Japan

Conducted research and development on computer vision for self-driving cars and ADAS.

  • Object detection
  • Efficient deep learning (model compression, transfer learning)
  • Image generation (GANs, diffusion models)
  • 3D model generation
  • Technology trends survey
 
 
 
 
 
Research Engineer
Mitsubishi Electric
Apr 2014 – Nov 2015 Hyogo, Japan

Developed software for driver monitoring system.

  • Driver’s cognitive distraction detection using deep neural networks
  • Occupant detection
  • AR HUD (Augmented Reality Head-Up Display)
3D Lighter: Learning to Generate Emissive Textures. SIGGRAPH Asia Posters, 2023.
BandRe: Rethinking Band-Pass Filters for Scale-Wise Object Detection Evaluation. International Conference on Machine Vision and Applications (MVA), 2023. Oral presentation, Honorable Mention Solution Award in Small Object Detection Challenge for Spotting Birds.
USB: Universal-Scale Object Detection Benchmark. British Machine Vision Conference (BMVC), 2022.
Domain Adaptation Regularization for Spectral Pruning. British Machine Vision Conference (BMVC), 2020. Acceptance rate: 29.1%.
Understanding the Effects of Pre-Training for Object Detectors via Eigenspectrum. International Conference on Computer Vision Workshops (ICCVW), 2019. Oral presentation, Acceptance rate: 9%, Best Paper Award Nominee.
Contrastive Relevance Propagation for Interpreting Predictions by a Single-Shot Object Detector. International Joint Conference on Neural Networks (IJCNN), 2019. Oral presentation, Acceptance rate: 36.4%.

Awards

  • President’s Commendation, Active Challenge Award
    DENSO, Dec 2022
  • CTO’s Commendation, Active Challenge Award
    DENSO, Dec 2022
  • Papers with Code Top Contributor Award
    Meta, Mar 2022
  • OpenMMLab Contributor of the Year Award 2021
    OpenMMLab, Jan 2022
  • The 56th Acacia Award
    Hiroshima University Senior High School, Jan 2007
  • The Award of the Minister of State for Science and Technology Policy at the 50th Japan Student Science Award
    Minister of Japan, Dec 2006
  • Outstanding Reviewer
    NeurIPS Datasets and Benchmarks Track 2022, BMVC 2022

Competitions

  • COCO test-dev AP 54.1% (to our knowledge, state-of-the-art accuracy by training within 24 epochs, as of November 14, 2020).
  • 1st place accuracy in 2nd AI Edge Contest.
  • Top single-stage detector in Waymo Open Dataset challenge 2D detection track, Workshop on Scalability in Autonomous Driving, CVPR 2020.
  • 1st place in NightOwls Detection Challenge 2020 object detection from single frame track, Workshop on Scalability in Autonomous Driving, CVPR 2020. Oral presentation.

Misc

Japanese translation

Survey and technical explanation

Unofficial demo

Reviewer

  • NeurIPS Datasets and Benchmarks Track (2022)
  • BMVC (2020, 2022)

Membership

  • Computer Vision Foundation (CVF)
  • Survey member of cvpaper.challenge