Reproduce “SpikeGS: 3D Gaussian Splatting from Spike Streams with High-Speed Camera Motion” 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

Jiyuan Zhang, Kang Chen, Shiyan Chen, Yajing Zheng, Tiejun Huang, Zhaofei Yu

SpikeGS introduces a framework that integrates 3D Gaussian Splatting with continuous spike camera streams, enabling high-fidelity 3D scene reconstruction and novel view synthesis with significantly faster training and rendering speeds under high-speed camera motion. It achieves over 10 dB PSNR improvement compared to prior methods and reduces training time from hours to minutes or seconds.

Key insights

  • Spike cameras provide blur-free, continuous data streams that are uniquely suited to overcome motion blur limitations of traditional cameras for high-speed 3D scene capture.
  • Combining the efficiency of 3D Gaussian Splatting's explicit scene representation with spike camera data can yield substantial improvements in both reconstruction quality and processing speed for novel view synthesis.
  • Both instantaneous (sharp, short-term) and aggregated (exposure-like, continuous) information from spike streams are valuable for robustly optimizing 3D scene parameters in an explicit representation like 3DGS.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.