Reproduce “TransGI: Real-Time Dynamic Global Illumination With Object-Centric Neural Transfer Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yijie Deng, Lei Han, Lu Fang
TransGI introduces an object-centric neural transfer model combined with an efficient probe-based lighting system to achieve real-time global illumination for dynamic scenes with complex materials. The system renders frames in approximately 6.7ms on an RTX 5000 GPU, enabling high-fidelity rendering with support for glossy materials and object-level dynamics.
Key insights
- Representing complex material properties in the frequency domain using spherical harmonics enables compact neural computation of transfer coefficients.
- Decoupling material representation via an object-centric neural model allows for high-quality, complex materials on dynamic objects without per-scene precomputation.
- An "across-probe radiance sharing" strategy using GPU atomic operations enables efficient high-resolution irradiance maps with fewer rays and improved parallelization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.