An efficient global illumination rendering method based on depth learning

Abstract

The invention discloses an efficient global light illumination rendering method based on depth learning, which comprises the following steps: 1) selecting or generating a plurality of groups of images, each group comprising k coarse color effect images which are calculated and rendered by using different photon collection radii for light illumination; For each group of images, k color effect images are stacked on three channels as the input of neural network. 2) training that input data by using the neural network to obtain a neural network model and parameter; 3) executing a photon mapping method according to the viewpoint parameters to be rendered and the three-dimensional scene, generating k color rough effect maps and stacking the k color maps on three channels as input data; Then thecurrent input data is processed by using the neural network model trained in step 2) and various parameters thereof to obtain the final composite rendering image. The invention only needs to illuminate the rough image with little light to synthesize the high-quality realistic rendering effect image.

Description

technical field [0001] The invention belongs to the field of computer graphics, and relates to a deep learning-based high-efficiency global illumination rendering method. Background technique [0002] Lighting calculation is a key issue in computer graphics. As games, movies, animation, and virtual reality are increasingly pursuing high realism in scene rendering, the speed of algorithms is increasingly demanding. However, the rendering algorithm based on global light illumination converges slowly, and often cannot meet the requirements of real-time rendering. [0003] As an advanced technology of machine learning, deep learning, combined with the development of high-parallel hardware, is showing great splendor in various fields of artificial intelligence. The deep learning theory is based on big data. Compared with the traditional artificial rule method, it has higher robustness and can handle various unexpected situations before the algorithm design. Among them, in the f...

AI Technical Summary

Problems solved by technology

However, this method may still have the following problems: a high-precision, high-robust neural network usually requires a deeper structure, more parameters, and a higher amount of calculation, so how to design an effective neural network under the premise of ensuring time efficiency? neural network to generate renderings with the same quality as traditional non-deep learning lighting calculation methods? Neural network is a data-based method. How much photon data can be used as input to get a general photon mapping-based global illumination algorithm?

Images