Translucent Material Parameter Estimation

Saip-Can Hasbay

Supervisor(s): David Hahn (TU Wien)

Vienna University

Abstract: In this paper, we present a workflow for optical material parameter estimation based on real-world images, numerical optimization, and differentiable rendering (using Mitsuba 3). We primarily focus on translucent materials and demonstrate our approach both for synthetic and real-world data. Specifically, we estimate parameters for two well-known material models: either a Principled BSDF (a surface model based on Burley’s Disney BSDF), or a Rough dielectric BSDF, describing volumetric homogeneous participating media. In order to solve the inverse rendering problem of acquiring suitable material properties from a (set of) given reference image(s), we present a software tool that handles the entire material reconstruction workflow. Furthermore, we also propose an experimental workflow to acquire the necessary reference images and potentially the 3D scene geometry. Our results show that our approach works on well-known materials such as acrylic glass, as well as newly designed materials, such as alginate, from experimental materials research.
Keywords: 3D Reconstruction, Physically-based Rendering, Rendering
Full text:
Year: 2023