- ② If the same Texture is bound to more than one component of a Material (e.g., to both PlasticMaterial::Kd and PlasticMaterial::Ks), the texture will be evaluated twice. This unnecessarily duplicated work may lead to a noticeable increase in rendering time if the Texture is itself computationally expensive. Modify the materials in pbrt to eliminate this problem. Measure the change in the system’s performance, both for standard scenes as well as for contrived scenes that exhibit this redundancy.
- ② Implement the artist-friendly “Disney BRDF” described by Burley (2012). You will need both a new Material implementation as well as a few new BxDFs. Render a variety of scenes using your implementation. How easy do you find it to match the visual appearance of existing pbrt scenes when replacing Materials in them with this one? How quickly can you dial in the parameters of this material to achieve a given desired appearance?
- ③ One form of aliasing that pbrt doesn’t try to eliminate is specular highlight aliasing. Glossy specular surfaces with high specular exponents, particularly if they have high curvature, are susceptible to aliasing as small changes in incident direction or surface position (and thus surface normal) may cause the highlight’s contribution to change substantially. Read Amanatides’s paper on this topic (Amanatides 1992) and extend pbrt to reduce specular aliasing, either using his technique or by developing your own. Most of the quantities needed to do the appropriate computations are already available— and in the SurfaceInteraction, and so on—although it will probably be necessary to extend the BxDF interface to provide more information about the roughness of any MicrofacetDistributions they have.
- ② Another approach to addressing specular highlight aliasing is to supersample the BSDF, evaluating it multiple times around the point being shaded. After reading the discussion of supersampling texture functions in Section 10.1, modify the BSDF::f() method to shift to a set of positions around the intersection point but within the pixel sampling rate around the intersection point and evaluate the BSDF at each one of them when the BSDF evaluation routines are called. (Be sure to account for the change in normal using its partial derivatives.) How well does this approach combat specular highlight aliasing?
- ③ Read some of the papers on filtering bump maps referenced in the “Further Reading” section of this chapter, choose one of the techniques described there, and implement it in pbrt. Show both the visual artifacts from bump map aliasing without the technique you implement as well as examples of how well your implementation addresses them.
- ③ Neyret (1996, 1998), Heitz and Neyret (2012), and Heitz et al. (2015) developed algorithms that take descriptions of complex shapes and their reflective properties and turn them into generalized reflection models at different resolutions, each with limited frequency content. The advantage of this representation is that it makes it easy to select an appropriate level of detail for an object based on its size on the screen, thus reducing aliasing. Read these papers and implement the algorithms described in them in pbrt. Show how they can be used to reduce geometric aliasing from detailed geometry, and extend them to address bump map aliasing.
- ③ Use the triangular face refinement infrastructure from the Loop subdivision surface implementation in Section 3.8 to implement displacement mapping in pbrt. The usual approach to displacement mapping is to finely tessellate the geometric shape and then to evaluate the displacement function at its vertices, moving each vertex the given distance along its normal. Refine each face of the mesh until, when projected onto the image, it is roughly the size of the separation between pixels. To do this, you will need to be able to estimate the image pixel-based length of an edge in the scene when it is projected onto the screen. Use the texturing infrastructure in Chapter 10 to evaluate displacement functions. See Patney et al. (2009) and Fisher et al. (2009) for discussion of issues related to avoiding cracks in the mesh due to adaptive tessellation.