Incorporating Lambertian Priors Into Surface Normals Measurement

The goal of photometric stereo is to measure the precise surface normal a 3D object from observations with various shading cues. However, non-Lambertian surfaces influence measurement accuracy due irregular Despite deep neural networks have been employed simulate performance surfaces, error in specularities, shadows, and crinkle regions hard be reduced. In order address this challenge, we here propose network that incorporates Lambertian priors better normal. paper, use initial under assumption as prior information refine measurement, instead solely applying observed cues deriving Our method utilizes reparameterize weights powerful fitting ability correct these errors caused by general reflectance properties. explorations include: (1) reduce learning hypothesis space, making our learn mapping same space improving learning, (2) provides differential features reconstruction details. Extensive experiments verify effectiveness proposed accurate on challenging benchmark dataset.