Reproduce “Learning Switchable Priors for Neural Image Compression” 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
Haotian Zhang, Yuqi Li, Li Li, Dong Liu
A framework for "universal photometric stereo" (UniPS) is presented, overcoming limitations of prior methods by operating without assumptions on physical lighting models. The UniPS-GLC network, trained on a new PS-Wild dataset, extracts global lighting contexts to recover surface normals under arbitrary and spatially-varying illumination, achieving superior accuracy compared to existing uncalibrated techniques.
Key insights
- Global lighting contexts, extracted by a neural network, can effectively encapsulate complex, spatially-varying illumination without explicit physical modeling.
- Training on a diverse, unconstrained dataset allows a photometric stereo network to generalize well to arbitrary and unseen lighting conditions.
- Decoupling the normal estimation from direct physical lighting models enhances robustness and versatility for real-world applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.