Language日本語English

Kaito Takahashi

高橋 海斗

I research how to recover the 3D motion of people from video. At hackathons and in my internship, I build apps with AI inside.

  • 3D motion reconstruction
  • Diffusion models
  • AI agent development
  • Mobile and web apps
Research demo3D motion from monocular video

Career

  1. Graduated2019.04 – 2024.03

    National Institute of Technology, Asahikawa College

    College of technology Department of Electrical and Computer Engineering

    asahikawa-nct.ac.jp

    2019.04
    Enrolled
    Spring 2023
    About a month of short-term study abroad at IUT de Lens, France IUT de Lens
    2024.03
    Graduated
  2. Ongoing2024.08 – present

    LANTERN Inc.

    Long-term internship

    lantern-inc.jp

    2024.08
    Started a long-term internship

    I work on websites and online stores (STUDIO, Shopify), LINE membership cards and points (LIFF, AWS), AI chatbots (Dify, Vercel), and automating SEO analysis and article production (Claude Code, MCP).

  3. Enrolled2025.04 – 2027.03 (expected)

    Kyoto Institute of Technology

    University Information Science, School of Science and Technology

    kit.ac.jp

    2025.04
    Transferred in as a third-year student
    2026
    Joined the Computer Vision Lab Lab
    2027.03
    Expected to graduate
  4. Planned2027.04 – (planned)

    Institute of Science Tokyo

    Graduate school Master's program, Department of Information and Communications Engineering, School of Engineering

    isct.ac.jp

    2027.04
    Planning to start a master's program (Kenji Suzuki Lab) Lab

Research

  1. Points on the body surface tracked in 2D, and the camera rotating slightly around the bodySTEP 1
  2. The 2D points and the rough 3D points from classical NRSfM, seen from the sideSTEP 2
  3. Diagram of a diffusion model that turns noise into 3D point trajectories, with each trajectory as one tokenSTEP 3
  4. Points in an image from a still camera, and their recovered 3D positions seen from the sideSTEP 4

3D motion reconstruction from monocular video without 3D ground truth

Computer Vision Lab, Kyoto Institute of Technology

This research aims to learn how things move in 3D from the 2D motion seen in video, without 3D ground truth such as motion capture. The learned model is then used to recover 3D motion from ordinary videos taken with a still camera.

  • PyTorch
  • NumPy
  • SciPy
  • CUDA

Read more

Works

Products I built with teams at hackathons.

Things I'm into

  • Into right now

    AION2

    An MMORPG by NCSOFT.

    AION2 official site

  • Played for years

    FPS

    I've played Apex Legends (battle royale) and VALORANT (5v5 tactical FPS).

    Apex LegendsVALORANT

  • Watched recently

    The Odyssey

    A film by Christopher Nolan (released in Japan in September 2026). Based on Homer's epic, it was shot entirely on IMAX.

    Official site

  • Making for fun

    Image generation

    I make images with Stable Diffusion (SDXL, ControlNet) and build tools for it with ComfyUI.

    Stable DiffusionComfyUI

Contact

For questions about my research or projects, please contact me by email.

kqito.inf@gmail.com