Short Bio

Kang Du (杜康) is a Ph.D. candidate in Computer Science at HKUST (GZ), advised by Prof. Zeyu Wang, and the Head of the Simulation Platform at Ant Lingbo (Ant Group). My research focuses on differentiable and inverse rendering for photorealistic simulation and light-aware world models, bridging computer graphics, vision, and robotics.

Experience

  • [Aug 2024] tech lead, Ant Group; PhD, HKUST-GZ
  • [Aug 2023] MBA, HKU
  • [Oct 2022] Project Manager, Tencent
  • [May 2021] Programmer Specialist, Meituan
  • [May 2020] Programmer, ByteDance
  • [May 2015] Founded a startup; BS, Texas A&M
  • Published on

    GS-ID: Illumination Decomposition on Gaussian Splatting

    ICCV2025

    We present GS-ID, a novel end-to-end framework that achieves comprehensive illumination decomposition by integrating adaptive light aggregation with diffusion-based material priors. Our method is designed to generate physically accurate 3D data with correct geometry, materials, and lighting — a critical foundation for training and evaluating world models and large language models that interact with or reason about the physical world.

    Read more →