Rendering global light transport in real-time using machine learning
Some implementations disclosed herein provide techniques and arrangements to render global light transport in real-time or near real-time. For example, in a pre-computation stage, a first computing device may render points of surfaces (e.g., using multiple light bounces and the like). Attributes for each of the points may be determined. A plurality of machine learning algorithms may be trained using particular attributes from the attributes. For example, a first machine learning algorithm may be trained using a first portion of the attributes and a second machine learning algorithm may be trained using a second portion of the attributes. The trained machine learning algorithms may be used by a second computing device to render components (e.g., diffuse and specular components) of indirect shading in real-time.
1. A method under control of one or more processors configured with a computer-readable memory device storing executable instructions, the method comprising:
rendering points on surfaces in a scene that includes global light transport;
determining attributes associated with each of the points, the attributes including at least a first portion of the attributes and a second portion of the attributes;
training a first point regression function based on the first portion of the attributes to create a first trained model, the first point regression function mapping the first portion of the attributes to at least a first component of indirect light; and
training a second point regression function based on the second portion of the attributes to create a second trained model, the second point regression function mapping the second portion of the attributes to at least a second component of indirect light.
2. The method of claim 1 , wherein the points on the surfaces in the scene are randomly selected.
3. The method of claim 1 , wherein:
the first portion of the attributes is associated with a specular component of the indirect light; and
the second portion of the attributes is associated with a diffuse component of the indirect light.
4. The method of claim 1 , wherein the first point regression function is based on a multiperceptron neural network algorithm.
5. The method of claim 1 , wherein the second point regression function is based on a random forests algorithm.
6. The method of claim 1 , wherein:
the first trained model is executable to determine diffuse shading values based on attributes of points on a surface; and
the second trained model is executable to render specular shading values based on the attributes of the points on the surface.
7. The method of claim 6 , wherein the diffuse shading values and the specular shading values are combined to create indirect shading values.
8. A computing device comprising:
one or more processors;
computer-readable memory device including instructions executable by the one or more processors to perform acts comprising:
rendering a plurality of points on one or more surfaces in a scene, the scene including light inter-reflections and light scattering;
determining attributes of the plurality of points;
training a first point regression function based on at least a first portion of the attributes;
training a second point regression function based on at least a second portion of the attributes; and
storing the first point regression function and the second point regression function for distribution to at least a second computing device.
9. The computing device of claim 8 , wherein training the first point regression function comprises training the first point regression function using a training set that includes diffuse shading components of the attributes.
10. The computing device of claim 8 , wherein training the second point regression function comprises training the second point regression function using a training set that includes specular shading components of the attributes.
11. The computing device of claim 8 , wherein the first point regression function is based on a multiperceptron neural network algorithm.
12. The computing device of claim 8 , wherein the second point regression function is based on a random forests algorithm.
13. The computing device of claim 8 , further comprising distributing the first point regression function and the second point regression function to the second computing device for rendering, substantially in real-time, indirect shading values of surfaces in the scene.
14. Computer-readable memory device including instructions executable by one or more processors to perform operations comprising:
determining a set of attributes for each of a plurality of points on one or more surfaces in a scene, the scene including illumination provided by global light transport, the global light transport including inter-reflections and scattering;
determining, based on the set of attributes, direct shading values;
determining, based on the set of attributes, diffuse lighting components by a first trained model comprising a first point regression function that maps the set of attributes to the diffuse lighting components;
determining, based on the set of attributes, specular lighting components by a second trained model comprising a second point regression function that maps the set of attributes to the specular lighting components;
combining the diffuse lighting components and the specular lighting components to create indirect shading values; and
combining the indirect shading values with the direct shading values to create full shading values for output to a display device.
15. The computer-readable memory device of claim 14 , wherein:
the first trained model comprises a first machine learning algorithm trained using a first set of training data; and
the second trained model comprises a second machine learning algorithm trained using a second set of training data.
16. The computer-readable memory device of claim 15 , wherein:
the first machine learning algorithm comprises a multiperceptron neural network algorithm; and
the first set of training data comprises a first portion of the set of attributes associated with a set of pre-computed pixels.
17. The computer-readable memory device of claim 16 , wherein:
the second machine learning algorithm comprises a random forests algorithm; and
the second set of training data comprises a second portion of the set of attributes associated with the set of pre-computed pixels.
18. The computer-readable memory device of claim 17 , wherein:
the first portion of the set of attributes comprises the diffuse lighting components; and
the second portion of the set of attributes comprises the specular lighting components.
19. The computer-readable memory device of claim 14 , wherein the set of attributes includes one or more of a position of a particular pixel, a normal associated with the particular pixel, a diffuse color associated with the particular pixel, a specular color associated with the particular pixel, or a view direction associated with the particular pixel.
20. The computer-readable memory device of claim 14 , wherein the diffuse lighting components and the specular lighting components are rendered substantially in real-time.