UNBIASED PATH TRACING OPTIMIZATION USING PARALLELISM, ADAPTIVE SAMPLING, AND RECONSTRUCTION

• Main Author: Rifqi Rahman, Raden
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84986
• Institution: Institut Teknologi Bandung

A major challenge in 3D rendering using the path tracing algorithm is achieving a high quality image with a low amount of noise in a reasonable amount of time. Therefore, it is necessary to incorporate optimization techniques to improve path tracing performance in terms of rendering time and image quality. Recent advances in path tracing optimization introduce optimization techniques that are capable of rendering a high quality image in a short time. However, such techniques often introduce a slight bias, which may divert the resulting image from the ground truth. Hence, unbiased path tracing optimization to boost rendering speed and image quality is yet to be explored. The state-of-the-art techniques for optimizing path tracer are adaptive sampling and recon- struction. In this final project, adaptive sampling and reconstruction techniques are leveraged to optimize an existing rendering system. To boost the path tracing performance even further, the paralellism technique using a graphics processing unit (GPU) is also used. Among existing path tracing systems, the PBRTv4 path tracer is selected. The selected adaptive sampling and reconstruction techniques are hierarchical stopping condition adaptive sam- pling and ray histogram fusion (RHF). In parallel rendering on the GPU, the optimization implementation uses the PBRTv4 wavefront architecture. The tests conducted against the optimization implementation show that parallelism, adaptive sampling, and reconstruction techniques are capable of improving path tracing performance. The optimized path tracer renders test images 1.61 to 12.12 times faster than the unoptimized path tracer. In terms of quality, the optimized path tracer renders test images with 3.8–25.9% higher structural similarity to the reference image than the unoptimized path tracer.