UCSD CSE 272 : Advanced Image Synthesis --- intro & simple BSDF part

UCSD CSE 272 : Advanced Image Synthesis --- intro & simple BSDF part

intro

course overview

smallpt : path tracing in 30 minutes

What is the color of a pixel?

reconstruction
path tracing
- intersection
- trace
  - diffuse relfection
  - mirror reflection
  - refraction
- lighting
- termination

what‘s more

lajolla

geometry primitive

struct Sphere : public ShapeBase {
    Vector3 position;
    Real radius;
};

struct TriangleMesh : public ShapeBase {
    /// TODO: make these portable to GPUs
    std::vector<Vector3> positions;
    std::vector<Vector3i> indices;
    std::vector<Vector3> normals;
    std::vector<Vector2> uvs;
    /// Below are used only when the mesh is associated with an area light
    Real total_area;
    /// For sampling a triangle based on its area
    TableDist1D triangle_sampler;
};

// To add more shapes, first create a struct for the shape, add it to the variant below,
// then implement all the relevant functions below.
using Shape = std::variant<Sphere, TriangleMesh>;

next event estimation

in addition to cosine-weighted hemisphere sampling, also sample a point on light

light sampling

  struct pdf_point_on_shape_op {
      Real operator()(const Sphere &sphere) const;
      Real operator()(const TriangleMesh &mesh) const;

      const PointAndNormal &point_on_shape;
      const Vector3 &ref_point;
  };

multi importance sampling

textures

用一阶Taylor展开来根本的解释mip map level原理
path tracing 怎么计算偏导数 ```cpp /// We adopt an approach for ray differentials /// that is used in Renderman. /// See Section 6.6 in "RenderMan: /// An Advanced Path Tracing Architecture for Movie Rendering" /// https://graphics.pixar.com/library/RendermanTog2018/paper.pdf /// The idea is to simplify Igehy's ray differential by only /// storing two quantities: a "radius" that describes positional /// differential, and a "spread" that describes directional differential.

/// For glossy/diffuse surfaces, Renderman used a heuristics based on the /// PDF of the sampling direction and use larger spread for low PDF. /// Here we use an even simpler heuristics: we linearly blend /// between the specular spread and a constant based on roughness. struct RayDifferential { // Radius is approximately (length(dp/dx) + length(dp/dy)) / 2 // Spread is approximately (length(dd/dx) + length(dd/dy)) / 2 // p is ray position, d is ray direction. Real radius = 0, spread = 0; // The units are pixels. }; ```