Reproduce “LUCES: A Dataset for Near-Field Point Light Source Photometric Stereo” 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
Roberto Mecca, Fotios Logothetis, Ignas Budvytis, Roberto Cipolla
The research introduces LUCES, the first real-world dataset specifically designed for near-field point light source photometric stereo, featuring 14 diverse objects with 52 images each, calibrated light parameters, and high-accuracy ground truth 3D models. This dataset enables standardized evaluation and development of algorithms capable of handling complex close-range illumination effects.
Key insights
- Far-field photometric stereo methods consistently failed under near-field conditions, empirically validating the necessity of algorithms specifically designed for close-range illumination.
- Deep learning-based photometric stereo methods generally surpassed classical optimization approaches on the LUCES dataset, indicating their effectiveness in learning complex non-linear shading relationships.
- Material specularity proved less detrimental to the performance of CNN-based methods on convex object regions, suggesting an implicit learning of complex BRDFs without explicit modeling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.