Three-dimensional (3D) rendering method and apparatus

A training method comprising:

• acquiring a first rendering result image by performing an entire rendering of a three-dimensional (3D) scene;
• determining some points from among a plurality of points in the 3D scene as sample points in the 3D scene for rendering the plurality of points of the 3D scene such that a density of the extracted sample points is greater for a greater diffuse reflection of the 3D scene, the determining of the sample points is performed by using a first neural network-based sample point determination model that uses texture information of feature information corresponding to the plurality of points of the 3D scene as an input;
• acquiring, by rendering the determined sample points, rendering result information corresponding to the sample points;
• acquiring a second rendering result image corresponding to an entire rendering of the plurality of points, the acquiring of the second rendering result image is performed by using a second neural network-based rendering image generation model provided the rendering result information of the determined sample points and the feature information corresponding to the plurality of points of the 3D scene as input;
• and
• adjusting, based on a result of a comparison between the first rendering result image and the second rendering result image, a parameter of one or both of the sample point determination model and the rendering image generation model,
• wherein the determined sample points include one or more points of each of plural 3D models, respectively, less than all points of each of the 3D models, in a three-dimensional space by coordinates, and the determined sample points correspond to a portion of vertices of the 3D models included in the 3D scene,
• wherein the adjusting of the parameter comprises adjusting a parameter of one or both of the sample point determination model and the rendering image generation model to reduce a loss, where the loss is determined based on the result of the comparison, a number of the sample points, and the density of the extracted sample points in the 3D scene.

A three-dimensional (3D) rendering apparatus comprising:

• at least one processor,
• wherein the at least one processor controls the 3D rendering apparatus to perform:
• extracting, from a 3D scene, some points from among a plurality of points in the 3D scene as sample points for rendering the plurality of points of the 3D scene such that a density of the extracted sample points is greater for a greater diffuse reflection of the 3D scene, based on texture information of the 3D scene;
• generating, by rendering the sample points, rendering result information corresponding to the sample points;
• and
• generating, based on the rendering result information and feature information corresponding to the plurality of points of the 3D scene, a rendering result image corresponding to an entire rendering of the plurality of points,
• wherein the extracted sample points include one or more points of each of plural 3D models, respectively, less than all points of each of the 3D models, in a three-dimensional space by coordinates, and the extracted sample points correspond to a portion of vertices of the 3D models included in the 3D scene,
• wherein the extracting of the sample points comprises extracting the sample points using a neural network-based sample point determination model configured to extract the sample points with at least one select density inferred by the sample point determination model based on feature information of the 3D scene provided to the sample point determination model.

A three-dimensional (3D) rendering apparatus comprises:

• a first processor configured to extract, from a 3D scene using a neural network, some points from among a plurality of points in the 3D scene as sample points for rendering the plurality of points of the 3D scene such that a density of the extracted sample points is greater for a greater diffuse reflection of the 3D scene, based on texture information of the 3D scene;
• and
• a second processor configured to generate, based on rendering result information corresponding to the sample points and feature information of the 3D scene, a rendering result image corresponding to an entire rendering of the plurality of points,
• wherein the extracted sample points include one or more points of each of plural 3D models, respectively, less than all points of each of the 3D models, in a three-dimensional space by coordinates, and the extracted sample points correspond to a portion of vertices of the 3D models included in the 3D scene,
• wherein the first processor is configured to estimate a first shading component for points other than the sample points using a neural network.