Reproduce “ICTPolarReal: A Polarized Reflection and Material Dataset of Real World Objects” 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

Jing Yang, Krithika Dharanikota, Emily Jia, Haiwei Chen, Yajie Zhao

A new dataset of real-world objects provides physically measured reflectance and material properties, captured using a polarized Light Stage. It delivers diffuse and specular separations, albedo, and surface normals for 218 objects, improving the performance of deep learning models in intrinsic image decomposition and relighting tasks.

Key insights

  • Polarization can be effectively exploited under controlled OLAT illumination to physically disentangle diffuse and specular reflections, providing crucial ground truth for intrinsic material properties.
  • Training deep learning models with real-world, polarization-separated reflectance data significantly improves their ability to generalize and perform accurate intrinsic image decomposition and relighting compared to synthetic data.
  • Generating normalized, specular-free multiview images through an inverse rendering model trained on this data can enhance the consistency and accuracy of downstream 3D reconstruction tasks.

Summary by alphaXiv.

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