Research On The Generation Method Of PBR Material

In recent years,the rendering effect of 3D models has become more and more realistic in both the game and the film industry.This is not only due to the rapid development of the graphics card industry,but also related to the development of graphics rendering technology.Physically Based Rendering(PBR)is a rendering technology recently developed.This technique simulates the representation of light in the real world to achieve the realistic rendering effect.Compared with the traditional rendering technology,PBR has been significantly improved.However,due to the complicated production process of the 3D model material texture required by the PBR and the high production threshold,the cost of the PBR model is much higher than the traditional model.Based on the research of PBR principle and related technologies,combined with the idea of deep learning,we propose a method for generating PBR material.Our method simplifies the production process of the PBR material and improves the production efficiency.The effectiveness and superiority of the proposed method are proved by observing the rendering effect.The main work of this paper is as follows:(1)A generation pipeline of PBR material is proposed,and the production process is divided into two stages.The first stage is to generate the albedo texture.The second stage is to predict the material type of the model based on the albedo texture and related information of the model to generate metallic texture,roughness texture and normal texture.(2)Before generating the albedo texture,it is necessary to extract the outline of the model.We use depth texture of the 3D model to assist the edge detection,which can better avoid the interference of environmental factors and generate the pure outline image of the model.(3)We propose a method to generate color information of the albedo texture based on Generative Adversarial Network.(4)In order to determine the type of material,we use the idea of classification,and train the classifier by synthesizing the data,which improved the prediction effect.And then,we generate the metallic texture,roughness texture and normal texture based on the material type.(5)A method of double-sided mapping and pixel expansion is proposed to solve the problem of generating texture integrity.(6)We design a PBR rendering system to observe the effect of our method.Compared with the traditional material,PBR material generated by our method has better visual performance.