| With the development of computer software and hardware technology,3D animation has been used in every aspects of life because of its simple,efficient,and expressive.In 1990 s,the academician Lu Ruqian,from Chinese Academy of Sciences,proposed a completely new animation production technology-the technology of the Full-Life Cycle Automatic Animation Generation.From the story of natural language until the generation of animation,every step is done with the aid of computer.In 2008,Zhang Songmao,Chinese Academy of Sciences researcher,proposed to apply the technology of the Full-Life Cycle Automatic Animation Generation to mobile phone text messages and realized it technically—the automatic generation system of mobile phones 3D animation(the generation system of animation).Machine learning has become the core method of the present artificial intelligence.All kinds of intelligent learning algorithms have been widely used in practical problems,such as prediction and evaluation.the automatic generation system of mobile phones 3D animation is based on a traditional knowledge method.it's facing two problems.Firstly,It lacks of learning ability,a large amount of accumulated animated data since 2008 has been wasted.Secondly,the quality of the animation can only depends on the ability of the system designer,and the user of the system can't participate in the animation generation.In order to solve these two problems,this paper designs and implements a active learning system for animation.The active learning system for animation can help the generation system of animation generate more satisfying animations for the users by constantly learning from experience.Concretely,according to the important role of the animation background scenes to the overall planning of the generation system,this paper mainly focuses on the selection of the background scenes of the animation.The main work of this paper includes:Firstly,this paper designs and implements the learning ability of the active learning system for animation.This paper design a random decision forests model for automatic animation generation to learn from the accumulated animated data.Basing on the design objectives and implementation principles of generation system of animation,this paper abstract the attributes and class of the model;After standardized processing a large number of historical data,we use the training sample set to get the random forest model;This paper also used experiments to optimize the parameters of random forest models helping the model direct the generation system of animation to generate more satisfied animations for users.Secondly,this paper designs and implements the active learning ability of the active learning system for animation.it constantly interacts with the generation system,the user and the manager,so that it can continuously receive samples from the generation system and automatically expand the number of training samples,analysis the state of the sample database and propose advise to the system manager about the need to update the learning model,all of this ensure the realization of the active learning ability.The active learning system for animation uses KFF algorithm based on the K-Means to select samples,and in order to solve the problem that the KFF don't consider the diversity of the samples,this system use statistical method to learn the novelty of the sample.In this paper,we determine the relative optimal parameters of random forest models through experiments,and prove the performance advantages of random forest models by comparing with decision tree models.And we statistics 614 tested messages from June 1,2017 to March 1,2018,and 379 SMS messages were finally rendered,including 369 SMS learning systems,which proved the feasibility of the learning system.By using the active learning system for animation,the probability of users' favorite scenes being picked up has been increased,and the probability that users do not like the scene being picked up has been reduced.As a whole,it achieve nearly 17% improvement of user satisfaction which proves the effectiveness of the active learning system for animation.For the first time,we design and implement a active learning system for animation that can active learning,it helps the generation system to select a user's favorite background scene after receiving SMS text,and improve user satisfaction of animation products.The active learning system for animation is closely connected with every step of the 3D animation automatic generation system,and the function of the active learning system for animation is limited to other steps.Further work includes the expansion of the learning model,the optimization of parameters,and the reduction of users,and the participation of the system managers. |
