Reproduce “Intrinsic Image Decomposition via Ordinal Shading” 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
Chris Careaga, Yağız Aksoy
A two-step deep learning pipeline enables high-resolution intrinsic image decomposition for 'in-the-wild' imagery by leveraging dense ordinal shading and a robust pseudo-ground truth generation strategy from multi-illumination real-world data, ensuring perfect image reconstruction. The method produces disentangled albedo and shading components with high detail and fidelity, effectively generalizing to diverse real-world scenes.
Key insights
- Predicting dense ordinal relationships of shading, rather than absolute values, simplifies the intrinsic image decomposition task for neural networks.
- Leveraging multi-resolution (global and local) ordinal information as input significantly improves decomposition quality and fine-grained detail.
- Generating robust pseudo-ground truth from real-world multi-illumination datasets effectively addresses the domain gap, enabling generalization to diverse in-the-wild scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.