Physics-based shading reconstruction for intrinsic image decomposition

Highlights：

• We are the first to use photometric invariance and deep learning to address the intrinsic image decomposition task.

• We propose albedo and shading gradient descriptors using physics-based models as novel priors.

• We show that a sparse shading map can be directly calculated from the corresponding RGB image gradients in a learning-free manner.

• We propose a novel deep learning model to leverage the physics-based shading map for the intrinsic image decomposition task.

• We are the first to directly address the color leakage problem in the estimated shading maps.