Papers & code
Centered on low-light image enhancement, we carry out our research in the open — publishing papers, code and evaluation results.
01Core Research
Low-light image enhancement
SD-Retinex: A Stable Retinex Method for Low-Light Image Enhancement Designed Around the Sigmoid Derivative(opens in a new tab)
A training-free, CPU-only low-light enhancement method built on the sigmoid derivative, with a pole-free real-axis kernel and a bounded tone map. Best PSNR and lowest color error among the training-free methods tested on LOL eval15.
複素指数カーネルに基づく高速・構造保持型 Retinex手法 XCRの提案(opens in a new tab)
A High-Speed and Structure-Preserving Retinex Method Based on a Complex Exponential Kernel: XCR
A Retinex method that estimates illumination with a separable kernel derived analytically from the complex exponential. Processes 1080p video at an average of 15.32 FPS on a CPU alone (measured on a Snapdragon X PC).
Fast Single-Scale Retinexの提案と低照度環境におけるリアルタイム骨格検出性能の評価(opens in a new tab)
Proposal of Fast Single-Scale Retinex and Evaluation of Real-Time Skeleton Detection Performance in Low-Light Environments
Proposes a fast Single-Scale Retinex and evaluates real-time skeleton detection in low-light environments.
02Publications
Team publications & talks
Re-Defining Vanishing Municipalities in Japan: A Multidimensional Clustering and SLM-Based Policy Insight Framework
A framework that combines multidimensional clustering with small language models (SLMs) to redefine “vanishing municipalities” in Japan and derive policy insights.
リアルタイム姿勢推定による弓道フォーム評価手法の提案とWeb実装
Proposal and Web Implementation of a Kyudo Form Evaluation Method Based on Real-Time Pose Estimation
Proposes a method for evaluating kyudo (Japanese archery) form using real-time pose estimation, implemented as a web application. Received the Outstanding Presentation Award.
03Roadmap
Research themes ahead
- 01
Dark-space inspection dataset
Capturing inspection footage in ship compartments and pipes while recording illuminance, distance and light output, with labels verified by people.
- 02
Enhancement meets detection
Pre-processing that adapts to darkness to stabilize object detection, anomaly detection and vision-language models.
- 03
Field-relevant evaluation
Measuring impact with metrics that matter on site, such as missed-defect rate and inspection time.
- 04
Faster at the edge
Real-time implementations for 1080p video, NPUs and older PCs.