Computer Graphics Forum (Proc. Eurographics Symposium on Rendering) 2023
Arvin Lin ¹ ² Yiming Lin ² Abhijeet Ghosh ¹ ²
Imperial College London ¹
Lumirithmic ²
Patent pending
Fig. 1: Acquired reflectance and transmission maps and surface geometry (a – f) of a liquor bottle, enabling realistic 3D rendering (g) under the Pisa lighting environment.
Abstract:
We present a practical method for acquisition of shape and plausible appearance of reflective and translucent objects for realistic rendering and relighting applications. Such objects are extremely challenging to scan with existing capture setups, and have previously required complex lightstage hardware emitting continuous illumination. We instead employ a practical capture setup consisting of a set of desktop LCD screens to illuminate such objects with piece-wise continuous illumination for acquisition. We employ phase-shifted sinusoidal illumination for novel estimation of high quality photometric normals and transmission vector along with diffuse-specular separated reflectance/transmission maps for realistic relighting. We further employ neural in-painting to fill gaps in our measurements caused by gaps in screen illumination, and a novel NeuS-based neural rendering that combines these shape and reflectance maps acquired from multiple viewpoints for high-quality 3D surface geometry reconstruction along with plausible realistic rendering of complex light transport in such objects.
Video Demonstrating Our Rendering Results:
Publication: Practical Acquisition of Shape and Plausible Appearance of Reflective and Translucent Objects. Arvin Lin, Yiming Lin, and Abhijeet Ghosh. Computer Graphics Forum (Proc. EGSR), 42(4), 2023.
