Reproduce “A CNN Based Approach for the Point-Light Photometric Stereo Problem” 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
Fotios Logothetis, Roberto Mecca, Ignas Budvytis, Roberto Cipolla
This research introduces a CNN-based iterative method that adapts deep learning photometric stereo to realistic point-light illumination by compensating for non-linear light attenuation and perspective viewing. The approach significantly outperforms state-of-the-art methods in reconstructing detailed 3D shapes from diverse materials and complex lighting, supported by the new LUCES dataset for near-field point-light scenarios.
Key insights
- Compensating for point-light attenuation is critical for adapting deep learning PS methods to near-field scenarios, drastically improving their performance compared to naive application.
- An iterative refinement process effectively converges to accurate depth and normal estimates, showing robustness to initial depth approximations.
- Perturbing training data with depth and light parameter noise enhances the network's generalization and robustness to real-world inaccuracies and miscalibrations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.