3D Material Rendering Using Generative Adversarial Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional rendering methods face challenges in achieving accurate and efficient rendering of complex 3D materials, with real-time rendering limited by performance and offline rendering being time-consuming.
Innovation Solution
A rendering method and device that utilize a set generative adversarial neural network to generate a 3D rendered graph from first original 3D information, improving accuracy and reducing calculation amounts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If real-time rendering is used for 3D materials, then rendering speed is improved, but rendering accuracy and quality deteriorate
Solution Approach 1:
The method performs preliminary actions by pre-training the generative adversarial network offline to learn the mapping between material properties and rendering appearances. During real-time rendering, the pre-trained network quickly generates high-quality material appearances without requiring complex ray tracing calculations, thus achieving both speed and accuracy
Solution Approach 2:
The generative adversarial network acts as an intermediary between the simple 3D model and the complex realistic rendering. It learns to map from simplified inputs to photorealistic outputs, bridging the gap between real-time performance requirements and high-quality rendering demands
2Manufacturing precision
If offline rendering with ray tracing is used for 3D materials, then rendering accuracy is improved, but rendering time increases
Solution Approach 1:
The method performs the computationally intensive learning process in advance during offline training. The generative adversarial network is trained offline using ray-traced images as ground truth, so that during actual rendering, the pre-trained model can quickly generate high-quality results without requiring time-consuming ray tracing executions
Solution Approach 2:
The generative adversarial network learns to copy the appearance characteristics of photorealistic ray-traced renderings. Instead of performing actual ray tracing during rendering, the network reproduces the visual effects by learning from pre-computed ray-traced examples, achieving similar quality with much lower computational cost during inference
Data Source
AI summary
A rendering method and device for a 3D material, an electronic apparatus, and a storage medium are provided. The method includes: acquiring first original 3D information of a 3D material to be rendered; generating an intermediate rendered graph according to the first original 3D information; and inputting the intermediate rendered graph into a generator of a set generative adversarial neural network, so as to obtain a 3D rendered graph.