Reproduce “NeRSP: Neural 3D Reconstruction for Reflective Objects with Sparse Polarized Images” 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
Yufei Han, Heng Guo, Koki Fukai, Hiroaki Santo, Boxin Shi, Fumio Okura, Zhanyu Ma, Yunpeng Jia
NeRSP is a neural 3D reconstruction method that accurately recovers the shape of reflective surfaces from sparse polarized images. It integrates both geometric and photometric cues derived from polarized images, achieving superior shape and normal reconstruction accuracy compared to prior methods on synthetic and real-world datasets using as few as six input views.
Key insights
- Jointly integrating geometric cues (surface normal azimuth from AoP) and photometric cues (polarimetric properties from Stokes vectors) significantly reduces ambiguity in shape estimation for reflective surfaces.
- The combination of these cues enables robust determination of 3D point positions and normals, effectively constraining the solution space even under sparse input view conditions.
- Utilizing polarized images provides direct geometric information that is less dependent on reflectance properties, allowing for better disentanglement of shape from radiance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.