Reproduce “DNA-Rendering: A Diverse Neural Actor Repository for High-Fidelity Human-centric Rendering” 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

Wei Cheng, Ruixiang Chen, Wanqi Yin, Siming Fan, Keyu Chen, Honglin He, Huiwen Luo, Zhongang Cai and 13 more

DNA-Rendering introduces a large-scale, high-fidelity dataset featuring diverse human performances, addressing limitations in existing human-centric rendering data by incorporating varied attributes and capture conditions. Benchmarking state-of-the-art methods on this dataset reveals current challenges in handling complex textures, non-rigid deformations, and human-object interactions, alongside demonstrating superior model generalization when trained on its comprehensive data.

Key insights

  • Benchmarking revealed that current rendering methods struggle significantly with complex texture rendering, non-rigid deformations, and accurate modeling of human-object interactions, especially for novel poses.
  • Methods incorporating human priors generally outperform category-agnostic approaches, and utilizing explicit appearance references aids generalization in novel identity rendering.
  • Training models on the diverse and high-fidelity DNA-Rendering dataset substantially improves cross-domain generalization robustness compared to training on less varied datasets.

Summary by alphaXiv.

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