New DensePol dataset enhances polarimetric vision with high-redundancy angle sampling

Researchers have introduced DensePol, a new dataset for polarimetric vision that utilizes a Division-of-Time (DoT) acquisition method to capture 180 full-resolution analyzer orientations at 1° intervals. This high-redundancy approach significantly improves polarization stability compared to existing Division-of-Focal-Plane (DoFP) datasets, reducing Angle of Line Polarization (AoLP) deviation. The dataset, comprising 2,018 paired RGB-polarization images, also includes fitting residuals and will be made publicly available with associated code. A novel deterministic diffusion-based RGB-to-polarization framework with cyclic AoLP representation and a local DoLP refiner has also been developed, demonstrating improved polarization prediction and downstream surface-normal estimation. AI

IMPACT This dataset and framework could advance AI's ability to interpret complex scene properties beyond RGB, potentially improving applications in robotics and computer vision.

RANK_REASON The item is a research paper detailing a new dataset and methodology for polarimetric vision. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• DensePol
• RGB color model

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →