Date of Award

5-19-2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

Holly Rushmeier

Abstract

This work explores the use of a B-spline-based approach for hyperspectral inverse rendering from RGB images, experimenting on both spectral and geometric reconstruction. While the B-spline method is less accurate than brute-force optimization, it offers significant improvements in computational efficiency- reducing both runtime and memory usage.

Our experiments show that the B-spline representation can approximate smooth spectral data effectively but struggles with sharper spectral features unless more knots are introduced. Notably, wavelengths near the edges of the visible spectrum (around 400 nm and 700 nm) were less stable during optimization, reflecting lower convergence reliability. Despite these challenges, the final RGB renders produced from B-spline-reconstructed spectra were visually accurate, indicating metameric equivalence to the ground-truth spectra.

These results position B-splines as a practical trade-off between accuracy and efficiency for spectral reconstruction in differentiable rendering pipelines, particularly for applications where real-time performance or resource constraints preclude brute-force methods.

Download

Share

COinS