Have you tested it on the Diligent dataset?

[yangyuze1]

How does it perform when you use the traditional least squares algorithm to estimate the normals of 96 images under different lighting conditions after removing the highlights? What is the MAE?