6.5 总结
基于体素化的烘焙器拥有超过BVH的光线追踪效率，可以更为快速地烘焙出Volume Lightmap和Visibility，并且由于不依赖硬件，因此非常易于扩展和引入前沿算法进行优化，易于接入不同的项目。不过，除这些优点外，还需要进行以下的优化才能更好地应用于更多的项目。
● 对于开放大世界来说，Voxelization的内存占用空间会很大，基于Voxelization的烘焙器采用的是内存换计算的策略：BVH只需要记录顶点和拓扑信息，Voxelization烘焙器还要记录每一个三角面内部的Voxel信息。内存数据量和场景大小成立方关系，超大场景的Voxel数据整体放进内存是不现实的，这里有两个解决的策略，一是对Voxel数据进行压缩，既包括对数据内容的压缩[21]，又包括对数据结构的压缩[22]；二是类似虚拟内存，对Voxel数据进行磁盘-内存间置换[23]。
● 针对超大场景，用3D Texture存储Volume Lightmap来采样GI同样有占用空间太大的问题，可以存储稀疏分布的Irradiance Probe，在Runtime时增加Alu计算进行插值，从而大大减少存储的数据量，使得数据量不和场景大小成立方关系。
● 不论是Volume Lightmap还是稀疏的Irradiance Probe，都有漏光、漏黑的问题，作者沿着采样方向进行了一定的偏移来缓解该问题。目前PC端业界的做法是对每一个Probe还要存储球面深度信息，如论文[24]，这样就可以大大减小漏光、漏黑的的程度，但该方法的缺点是显而易见的，会引入大量的数据，并且在Runtime时Alu的开销会大大增加，所以该方法要在手机端使用还需要进一步的优化。
[1] SCHWARZ M, SEIDEL H-P. Fast parallel surface and solid voxelization on GPUs[J]. ACM transactions on graphics (TOG), 2010, 29(6): 1-10.
[2] KARRAS T.Maximizing parallelism in the construction of BVHs,octrees, and k-d trees[C]//Proceedings of the Fourth ACM SIGGRAPH/Eurographics conference on High-Performance Graphics, 2012: 33-37.
[3] HARADA T, HOWES L. Introduction to GPU radix sort-Heterogeneous Computing with OpenCL[M]. San Mateo, CA. Morgan Kaufman, 2011.
[4] LAINE S, KARRAS T.Efficient sparse voxel octrees[J]. IEEE Transactions on Visualization and Computer Graphics, 2010, 17(8): 1048-1059.
[5] ENDERTON E, SINTORN E, SHIRLEY P, et al. Stochastic transparency[J]. IEEE transactions on visualization and computer graphics, 2010, 17(8): 1036-1047.
[6] PHARR M, JAKOB W, HUMPHREYS G.Physically based rendering: From theory to implementation[M]. Morgan Kaufmann, San Mateo, CA, 2016.
[7] HILLAIRE S, ROUSIERS C de, APERS D. Real-Time Raytracing for Interactive Global Illumination Workflows in Frostbite[C]//Game Developers Conference, 2017.
[8] AILA T, LAINE S.Understanding the efficiency of ray traversal on GPUs[C]//Proceedings of the conference on high performance graphics, 2009: 145-149.
[9] YOUNG E. Direct Compute Optimizations and Best Practices[C]//GPU Technology Conference, 2010.
[10] RONG G, TAN T-S. Variants of jump flooding algorithm for computing discrete Voronoi diagrams[C]//4th International Symposium on Voronoi Diagrams in Science and Engineering (ISVD 2007). IEEE, 2007: 176-181.
[11] SLOAN P-P, KAUTZ J, SNYDER J.Precomputed radiance transfer for real-time rendering in dynamic, low-frequency lighting environments[C]//Proceedings of the 29th annual conference on Computer graphics and interactive techniques, 2002.
[12] IWANICKI M, SLOAN P-P.Precomputed lighting in Call of Duty:Infinite Warfare[C]//SIGGRAPH 2017 Course: Advances in Real-Time Rendering in Games, 2017.
[13] SLOAN P-P. Stupid spherical harmonics (sh) tricks[C]//Game developers conference, 2008.
[14] MEURER A, SMITH C P,PAPROCKI M. SymPy:symbolic computing in Python[J].PeerJ Computer Science, 2017, 3:e103.
[15] PRESS W H, WILLIAM H, TEUKOLSKY S A, et al. Numerical recipes 3rd edition:The art of scientific computing[M]. Cambridge university press, UK, 2007.
[16] KAVAN L, BARGTEIL A W, SLOAN P-P. Least squares vertex baking[C]//Computer Graphics Forum.Wiley Online Library, 2011: 1319-1326.
[17] JIMÉNEZ J, WU X,PESCE A.Practical real-time strategies for accurate indirect occlusion[C]//SIGGRAPH 2016 Courses:Physically Based Shading in Theory and Practice, 2016.
[18] EL GARAWANY R. Deferred lighting in uncharted 4[C]//SIGGRAPH 2016 Courses: Advances in Real-Time Rendering in Games, 2016.
[19] MAZONKA O. Solid angle of conical surfaces,polyhedral cones, and intersecting spherical caps[OL].arXiv preprint arXiv:1205.1396, 2012.
[20] OAT C, SANDER P V. Ambient aperture lighting[C]//Proceedings of the 2007 Symposium on interactive 3D Graphics and Games, 2007:61-64.
[21] DOLONIUS D, SINTORN E, KÄMPE V, et al. Compressing color data for voxelized surface geometry[J]. IEEE transactions on visualization and computer graphics, 2017, 25(2): 1270-1282.
[22] VILLANUEVA A J,MARTON F,GOBBETTI E. Symmetry-aware Sparse Voxel DAGs (SSVDAGs) for compression-domain tracing of high-resolution geometric scenes[J].Journal of Computer Graphics Techniques, 2017,6(2): 1-30.
[23] BAERT J, LAGAE A, DUTRÉ P.Out-of-core construction of sparse voxel octrees[C]//Proceedings of the 5th high-performance graphics conference, 2013: 27-32.
[24] MAJERCIK Z,GUERTIN J-P,NOWROUZEZAHRAI D, et al. Dynamic diffuse global illumination with ray-traced irradiance fields[C]//Journal of Computer Graphics Techniques, 2019.