Reproduce “OutCast: Outdoor Single-image Relighting with Cast Shadows” 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
David Griffiths, Tobias Ritschel, Julien Philip
Researchers from Adobe Research and University College London developed OutCast, a system for photorealistic outdoor image relighting that accurately generates and removes cast shadows from a single RGB input image. The method employs a learned image-space ray-marching layer, achieving superior quantitative and qualitative results compared to prior single-image techniques.
Key insights
- Noisy and incomplete single-image depth maps can be robustly utilized for generating complex cast shadows when intelligently compensated for by a learned network.
- Explicitly modeling physics-inspired operations, like ray-marching, within a deep learning framework effectively addresses the ill-posed problem of inferring 3D occluding geometry from a single 2D image.
- Training with a mix of ground truth and estimated depth helps the network learn fundamental geometric principles while developing robustness to real-world depth inaccuracies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.