Reproduce “4K4D: Real-Time 4D View Synthesis at 4K Resolution” 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

Zhen Xu, Sida Peng, Haotong Lin, Guangzhao He, Jiaming Sun, Yujun Shen, Hujun Bao, Xiaowei Zhou

4K4D introduces a method for real-time 4D view synthesis, generating high-fidelity views of dynamic 3D scenes at 4K resolution. The approach achieves over 80 FPS at 4K and 400 FPS at 1080p on commodity hardware, significantly outperforming previous real-time dynamic neural rendering techniques.

Key insights

  • Explicit, hardware-friendly 4D point cloud representations are highly effective for overcoming the speed bottlenecks of implicit neural rendering in dynamic scenes while maintaining high visual quality.
  • The strategic combination of discrete image-based rendering and continuous spherical harmonics in a hybrid appearance model effectively balances expressiveness, memory footprint, and computational efficiency.
  • Integrating traditional graphics hardware rasterization via differentiable depth peeling provides crucial speed gains, allowing neural scene representations to achieve truly real-time performance at demanding resolutions like 4K.

Summary by alphaXiv.

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