Reproduce “Robust Low-light Scene Restoration via Illumination Transition” 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
Ze Li, Feng Zhang, Xiatian Zhu, Meng Zhang, Yanghong Zhou, P. Y. Mok
RoSe, a framework developed at HKUST and HKPU, robustly synthesizes normal-light novel views from noisy, low-light multiview images by modeling illumination as a 3D transition field. It achieved a mean PSNR of 26.28, SSIM of 0.808, and LPIPS of 0.379 on the LOM dataset, outperforming existing state-of-the-art methods.
Key insights
- Low-light scene restoration can be effectively framed as an illuminance transition estimation problem in 3D space, bypassing traditional ill-posed image decomposition.
- Ensuring multiview consistency for illumination enhancement is crucial, achieved by making the illuminance transition field dependent solely on 3D spatial location.
- The spatially smooth, low-rank property of illumination can be leveraged to denoise effectively without explicit noise modeling, preventing noise from being learned as scene geometry.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.