The Research Of Time Prediction For Rendering Based On PSO-SVR

Rendering refers to the progress of generating digital images or bitmap images from models using three-dimensional software, these three-dimensional software including Maya, 3ds Max, and so on. The rendering model is described by a strict definition of the language or data structure for 3D objects, which contains geometry, viewpoint, texture, and illumination information. Rendering time refers to the time spent on the process of generating images from models. The length of the rendering time is related to many factors, such as geometric complexity, resolution, material, etc. At present, a lot of rendering company according to the principle of first come first served assign rendering tasks to nodes in the cluster, this principle sometimes causes the waste of the cluster resource. At present, the urgent problem of rendering company is how to make a reasonable and effective scheduling strategy.In view of the above problems, in this paper we through in-depth to analyze the rendering system which used V-Ray renderer and extract parameters that affect rendering time, and combine with machine learning methods,we study rendering time prediction method based on support vector regression optimized by particle swarm optimization to estimate the rendering time.According to the rendering time we can make a reasonable cluster scheduling strategy, so we can make use of the resources effectively. At the same time,the rendering time can provide a reference for setting up the price.The main work of this paper is as follows:1.We analyze the rendering steps and the specific implementation process in detail,and from the resolution, geometric complexity degree, material and light to research rendering system which used V-Ray renderer. Through experiments we extract thirteen parameters which affect the rendering time,there are the resolution, geometry, material quantity, refraction or reflection times, subdivision of lights, light quantity, adapting the number, the noise threshold, global subdivision multiplier, irradiance map presupposition, hemisphere subdivision, maximum subdivision and minimum subdivision.These factors provide sample parameters for this method which this paper researches.2.In depth study of the support vector regression machine theory foundation, due to the training parameters of support vector machine regression have an impact on prediction performance, the better parameters combination is needed in order to obtain better performance of support vector regression. Therefore,we introduce an advanced particle swarm optimization algorithm(PSO). We adopt random search strategy of particle swarm optimization algorithm to optimize the training parameters of support vector regression, such as the kernel parameters, the penalty factor and the insensitive loss function and construct a time prediction model based on PSO-SVR.3.Through specific examples of analysis, the rendering time prediction model based on PSO-SVR practical application effect for verification, compared with the BP neural network and the stepwise regression. The experimental results show that in time prediction of rendering PSO-SVR is much better than BP neural network and stepwise regression, which prediction accuracy is greatly improved, and have good generalization ability.