Neural Reflectance Capture in the View-Illumination Domain.

Area of Science:

• Computer Vision
• Computer Graphics
• Material Science

Background:

• Accurately capturing surface reflectance properties of 3D objects is crucial for realistic rendering and material analysis.
• Existing methods often struggle with efficiency, complex lighting conditions, and unknown reflectance properties on non-planar surfaces.

Purpose of the Study:

• To develop a novel and efficient framework for capturing unknown anisotropic reflectance on non-planar 3D objects.
• To leverage multi-view photometric measurements and deep learning for robust reflectance reconstruction.

Main Methods:

• A deep neural network is designed to exploit multi-view coherence for efficient information aggregation.
• The framework utilizes a high-performance illumination multiplexing setup to probe the 4D view-lighting domain.
• Photometric measurements under learned lighting patterns from multiple views serve as input for the network.

Main Results:

• The proposed framework successfully reconstructs anisotropic reflectance with high fidelity on various physical objects.
• The system demonstrates superior acquisition efficiency compared to current state-of-the-art methods.
• An evaluation of sampling parameter impacts on network performance is presented.

Conclusions:

• The novel framework offers an efficient and effective solution for capturing unknown object reflectance.
• The deep learning approach, combined with optimized illumination, significantly advances 3D object material property acquisition.
• This work provides a foundation for more sophisticated material capture and analysis techniques.