Reproduce “SeaSplat: Representing Underwater Scenes with 3D Gaussian Splatting and a Physically Grounded Image Formation 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
Daniel Yang, John J. Leonard, Yogesh Girdhar
SeaSplat integrates a physically grounded underwater image formation model with 3D Gaussian Splatting to enable real-time 3D reconstruction and novel-view synthesis of underwater scenes. This approach yields enhanced geometric fidelity and true color restoration while significantly outperforming prior NeRF-based methods in computational efficiency, such as a 850x speedup in rendering time.
Key insights
- Integrating a physical underwater image formation model with an explicit 3D representation like Gaussian Splatting effectively disentangles medium effects from scene content, leading to more accurate geometry and color restoration.
- Carefully designed regularization losses, informed by principles of underwater imaging and color constancy, are crucial for guiding the optimization towards physically plausible solutions and preventing common artifacts like 'floaters'.
- Learning scene true colors and medium parameters simultaneously within a computationally efficient framework enables real-time novel-view synthesis and true color restoration, overcoming a major limitation of prior implicit methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.