雙向路徑追蹤是一種基於真實世界中的物理現象來繪製高品質影像的演算法。有別於一般的路徑追蹤，雙向路徑追蹤同時考慮到了多種重要性採樣策略來加快繪製速度，但這個演算法仍然要使用大量的運算資源，並不能即時繪製。隨著GPU的進步，GPU不只支援平行運算，還有特別的運算單元來對光線追蹤運算進行硬體加速。本文討論了雙向路徑追蹤的理論以及如何在Nvidia OptiX 7框架上建立Bidirectional Path Tracer。為了比較雙向路徑追蹤與一般的路徑追蹤，本文分別使用兩種演算法，在限定的時間內繪製特定的場景並計算結果與參考圖的平方差。實驗結果顯示，在實驗的設置下，雙向路徑追蹤確實優於一般的路徑追蹤。

Bidirectional path tracing is a physical base rendering algorithm that generates rays following physical rules to render photorealistic images. Unlike simple path tracing, bidirectional path tracing concerns more than one importance sampling strategies to reduce render time, but this algorithm still cost a lot of computation resources which means it can only be used in offline rendering. Fortunately, GPUs have become more and more powerful today. GPUs support not only parallel computing but hardware acceleration to perform ray tracing operation. In this paper, we discuss how bidirectional path tracing works and build a bidirectional path tracer on Nvidia OptiX 7 framework. In order to compare between bidirectional path tracer and simple path tracer, we render specific scenes by both algorithms in a limit time and calculate squared error. The result shows that the performance of bidirectional path tracing algorithm is better than simple path tracer.

[1] NVIDIA OptiX 7.5 –Programming Guide, https://raytracingdocs.nvidia.com/optix7/guide/index.html

[2] Veach, E. (1998). Robust Monte Carlo methods for light transport simulation. Stanford University.

[3] Lafortune, E. P., AND Willems, Y. D. (1993). Bi-directional path tracing.

[4] Kollig, T., AND Keller, A. (2002). Efficient bidirectional path tracing by randomized quasi-Monte Carlo integration. In Monte Carlo and Quasi-Monte Carlo Methods
2000 (pp. 290-305). Springer, Berlin, Heidelberg.