ELECTRONIC DEVICE AND METHOD OF SCENE RENDERING

Smart overview of the Invention

The invention involves an electronic device equipped with a processor and memory that stores instructions for rendering scenes. This process utilizes machine learning models to select a 3D point in a scene from a specific viewpoint. Based on this selection, the device calculates partial values of the bidirectional scattering distribution function (BSDF) using models connected to each other, which are then combined to form a full BSDF value. This value, along with ray direction and probability density function (PDF), is used to generate an image of the scene.

Machine Learning Models

The rendering process employs two interconnected machine learning models. The first model helps in obtaining a partial BSDF value using view direction, ray direction, and material information. The second model, which receives inputs from an intermediate layer of the first model, calculates another partial BSDF value. Both models can include multi-layer perceptron (MLP) layers, and their combined outputs determine the full BSDF value for rendering.

Ray Direction and Lobes

The method involves sampling the ray direction based on multiple lobes, which correspond to potential reflections or transmissions at the 3D point. A third machine learning model is used to estimate spherical Gaussian (SG) parameters for these lobes, defining attributes like axis, sharpness, and amplitude. A PDF is determined based on these SG distributions, guiding the sampling of ray directions for realistic image synthesis.

Handling Transparent Materials

When the scene includes objects with transparent materials, the device generates a frame image within a specified time by calculating pixel values using the estimated BSDF value for rays passing through these objects. This ensures efficient rendering of scenes with complex material interactions, maintaining realism and computational efficiency.

Flexibility and Modifications

The description emphasizes the flexibility of the system, allowing modifications and alterations without departing from the core invention. The terms used are meant to distinguish components without limiting their functionality, ensuring the system can adapt to various configurations and applications within the scope of the disclosure.