Practice realistic human skin rendering in renderman Production Pipeline

Although there are already advanced skin SSS solutions based on spectral and multilayer materials, including a spectral bssrdf for shading human skin and the latest a layered, heterogeneous Reflectance Model for acquiring and rendering human skin has outstanding results. However, compared with actual CG applications, it is the ultimate R & D goal to allow artists to quickly and accurately implement the desired results in their minds, and most importantly, the objects presented in CG production do not necessarily exist in reality, such as the Hulk, which leads to a highly nonlinear complex processing model, it cannot be adjusted from the actual situation. Therefore, for practical applications, the Single Dipole diffusion model described in a rapid hierarchical Rendering Technique for translucent materials can be directly controlled due to the low number of parameters, I think it is a good application-level solution. Currently, Pixar renderman 13.5 and blender have their own implementations. Note that the fast_sss series shader provided in Mental Ray for Maya divides SSS into forward and backward scattering components, which can simulate SSS effects after simple linear mixing, and the effect is controllable, but it is a good choice.

To provide an accurate reference to artists, I have referred to the data in the Merl/ETH face scan database and converted all its data from the MATLAB matrix format to the XML format. This database has a total of 156 samples, covering the coefficient data of albedo and Torrance-Sparrow BRDF in 10 parts of human faces of various skin colors and different ages. Here is the XML version of the compressed package download. For more information, see here. Before use, please follow the Merl/ETH copyright statement for non-commercial educational research purposes only.

The entire process can be summarized in this way. First, the desired effect is determined by the highest artistic guidance, and TD and the texture artist follow the concept, guided by TD and corrected the numerical distribution of the diffuse diagram drawn by the artist, the dmfp and BRDF coefficients required by each part of the TD experiment are provided to the artist for drawing, the shader also provides sufficient parameters for independent correction of the distribution of each component. Diffuse mean free path (dmfp) is also the coefficient to be solved by TD. After bake is imported into the point cloud of renderman, it is calculated by ptfilter. The dmfp coefficient has a huge impact on the effect. The following three groups are the adjustment results of different dmfp components.

The following is a simple human face model that I created using the existing model and practiced the entire process, including testing the M and \ rho_s and dmfp required by ts, manually draw a fp32 texture in PS. However, given that the filter conditions are not able to correct the color at a later stage, the effect on different display devices varies greatly, therefore, there is a huge gap between the actual product-level quality. I am writing a complete documentation in latex format, so stay tuned.