Reproduce “Deblurring 3D 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
Byeonghyeon Lee, Howoong Lee, Xiangyu Sun, Usman Ali, Eunbyung Park
Researchers at Sungkyunkwan University introduce a framework for deblurring 3D Gaussian Splatting, enabling real-time rendering of sharp 3D scenes from blurry input images. This method achieves state-of-the-art deblurring performance, with rendering speeds exceeding 800 frames per second, on various real and synthetic blur datasets.
Key insights
- Spatially varying blur can be effectively modeled by dynamically adjusting the scale, rotation, and position of individual 3D Gaussians through an auxiliary neural network.
- Decoupling the blur modeling process (active MLP during training) from the sharp rendering process (deactivated MLP during inference) is crucial for maintaining real-time performance.
- Addressing the sparsity of initial point clouds from blurry SfM inputs, through targeted densification and depth-aware pruning, is essential for robust high-fidelity 3D scene reconstruction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.