Hybrid Quantum-Inspired RGB Patch Reconstruction Using Mitsuba 3 and Qiskit

Abstract

        Path tracing produces physically plausible images by estimating light transport through Monte Carlo sampling, but high-quality renders require many samples per pixel. This paper presents a modest hybrid quantum-classical benchmark that connects physically based rendering in Mitsuba 3 with quantum-circuit experiments in Qiskit. A Cornell Box scene is rendered classically, an 8 x 8 RGB region of interest is extracted from a 256 samples-per-pixel reference image, and each red, green, and blue channel value is encoded as a one-qubit measurement probability using an Ry rotation. Two estimators are evaluated: direct Bernoulli sampling of the prepared qubit and a Grover-style amplified circuit followed by probability inversion. The naive estimator behaves as expected, with RGB MSE decreasing from 0.00480 at 32 shots to 0.000124 at 1024 shots. In contrast, the current amplified estimator remains near 0.036 RGB MSE across shot counts, indicating that the tested amplification circuit or inversion is not suitable for the present scalar-channel reconstruction task. The work therefore contributes to a reproducible Mitsuba-Qiskit benchmark and a diagnostic negative result, rather than evidence of full quantum ray tracing or practical rendering speedup.