Skip to content
Toma Okugawa
  • Research
  • Projects
  • Publications
  • Recognition
  • Updates
  • About
日本語

Research

Research

Training-free low-light image enhancement, on-device pose estimation, and small language models for regional data.

6

  • Four rows of LOL eval15 scenes compared across seven columns: the low-light input, then the SSR, MSR, LIME, XCR and SD-Retinex outputs, and the ground truth on the right

    SD-Retinex: A Stable Retinex Method for Low-Light Image Enhancement Designed Around the Sigmoid Derivative

    Training-free Retinex enhancement that moves the sigmoid derivative to the real axis: 16.47 dB PSNR on LOL eval15, about 54 FPS at 600×400 on one CPU thread

    • 27 Jun 2026
    • Zenodo
    • Preprint
  • A High-Speed and Structure-Preserving Retinex Method Based on a Complex Exponential Kernel: XCR

    A training-free Retinex method whose off-center kernel, derived from a complex exponential, beats MSR and a LIME-based method on SSIM at 15 FPS (1080p) on a CPU

    • 20 Nov 2025
    • Jxiv
    • Preprint
  • Proposal of Fast Single-Scale Retinex and Evaluation of Real-Time Human Pose Estimation Performance under Low-Light Conditions

    A luminance-only Retinex with a separable Gaussian, about 5.7× faster than 2-D SSR, that lets MediaPipe Hands find landmarks in near-dark video

    • 12 Nov 2025
    • Jxiv
    • Preprint
  • Proposal and Web Implementation of a Kyudo Form Evaluation Method Based on Real-Time Pose Estimation

    Real-time pose estimation that detects kai, the full draw, and scores kyudo form from joint positions; Outstanding Presentation Award at SJCIEE 2025

    • 22 Sept 2025
    • Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers (SJCIEE 2025)
    • Published
  • Re-Defining Vanishing Municipalities in Japan

    Clustering national statistics with UMAP and fine-tuning a locally run small language model to describe vanishing municipalities by more than population

    • Oct 2026
    • IEEE 15th Global Conference on Consumer Electronics (GCCE 2026)
    • To appear
  • Recognition Without Revenue: Designing a Failure-Case-Based Entrepreneurship Education Model for Japanese KOSEN Students

    A design study that turns Second, a student venture closed with zero revenue, into a teaching case and a five-year entrepreneurship education model for KOSEN

    • 12 Sept 2026
    • EdArXiv
    • Preprint

Toma Okugawa

Making cameras see in the dark, and phones understand human motion.

i24010@yuge.kosen-ac.jp
  • Research
  • Projects
  • Publications
  • Recognition
  • Updates
  • About
  • GitHub
  • researchmap
  • ORCID
  • LinkedIn
  • Medium
  • J-GLOBAL

© 2026 Toma OkugawaBuilt with Astro. Hosted on Cloudflare Pages.SourceBack to top ↑