Reproduce “Gaussian in the Dark: Real-Time View Synthesis From Inconsistent Dark Images Using Gaussian Splatting” 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
Sheng Ye, Zhen-Hui Dong, Yubin Hu, Yu-Hui Wen, Yong-Jin Liu
Gaussian-DK introduces a 3D Gaussian Splatting-based method for high-quality, real-time novel view synthesis from multi-view inconsistent images captured in dark environments. It achieves this by modeling camera response to manage brightness variations and employing a gradient scaling strategy to mitigate rendering artifacts.
Key insights
- Explicitly disentangling the consistent physical radiance field from the camera's view-dependent imaging process is crucial for handling multi-view inconsistencies in dark environments.
- Combining global exposure parameters with local, learnable Gaussian light features allows for precise control over brightness effects and effective compensation for varied scene illumination.
- Strategically scaling gradients based on Gaussian distance to the camera during training provides an effective mechanism to prevent 'floater' artifacts without compromising high-frequency scene details.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.