基于径向基函数的实时烟雾渲染

RBF-based real time smoke rendering

即使使用当前最先进的并行计算设备,包含散射的高分辨率烟雾渲染依然是一个巨大的挑战。补偿光线步进法通过预处理将烟雾仿真数据分解为一组径向基函数的近似值和残差项。在只考虑径向基函数近似的密度下计算源辐射分布,并在光线步进过程中加入残差的影响。虽然这种算法可以带来实时高质量的渲染结果,但是随着烟雾密度数据分辨率的提升预处理时间很快上升到不可接受的程度。为解决这一问题,我们引入多分辨率方法。我们的算法为由原数据构造从低到高不同层次分辨率的数据,逐层指定更多径向基函数进行预处理,使用当前层结果为后一层处理做初始化。此外还引入了概率控制防止可能的长时间不收敛情况,并借鉴遗传变异思想预防优化过程陷入局部最优解。实验证实,我们的算法节约了50%以上的计算时间。

Even with the state-of-the-art parallel computing devices,real-time volumetric rendering of smoke data is still challenging especially when scattering is involved and higher resolution dataset is used.Compensated ray marching algorithm is based on a decomposition of the input smoke animation into a set of radial basis functions(RBFs)and a residual field.The source radiance distribution within the smoke is computed from the RBF approximation of the density field and residual field is compensated back into the radiance integral during ray marching.While this algorithm can bring in high quality rendering result,the preprocessing time would be prohibitive for high resolution smoke density data.We introduce multi-resolution method to solve this problem.Such an algorithm starts by constructing a hierarchy of resolutions from the original data,and then progressively fit the smoke data of each resolution with increasing numbers of RBF kernels in a coarse-to-fine manner.We also apply a probability controller to prevent potential long time non-convergence,and reference heritable variation idea to prevent the optimization from trapping in locally optimal solution.This algorithm leads to reduction of precomputation time by 50%.