Neural Network Animation Generation Reducing Memory Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing animation data generation technologies require large internal memory for storing massive data, leading to poor query performance and limiting the development of motion matching technology in animation engines, which affects the user experience in video games.
Innovation Solution
The use of a pre-trained neural network to generate animation data by increasing the feature dimension of a virtual object based on its trajectory and bone features, reducing the need for massive data storage and improving query performance by only requiring storage of weight data related to the neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If motion matching technology is used to generate animation data, then animation quality is improved, but internal memory occupation increases significantly
Solution Approach 1:
The patent extracts only the essential features (trajectory features and bone features) from the virtual object's running data, rather than storing and processing complete animation datasets. This extraction approach maintains animation quality while significantly reducing memory requirements by focusing only on the critical motion parameters needed for generation.
Solution Approach 2:
The patent replaces the traditional mechanical motion matching system (which requires storing massive animation databases in internal memory) with a neural network-based system. The neural network learns motion patterns during training and generates animations computationally during runtime, substituting the memory-intensive mechanical lookup approach with a more efficient learning-based generation approach.
2Adaptability or versatility
If massive animation data is stored in internal memory for motion matching, then animation variety is improved, but query performance deteriorates
Solution Approach 1:
The patent performs preliminary action by training the neural network offline before runtime. During training, the network learns from diverse animation data to acquire broad motion knowledge. At runtime, it rapidly generates varied animations using this pre-acquired knowledge, eliminating the need for real-time querying of large databases and thus improving query performance while maintaining animation variety.
Solution Approach 2:
The patent changes the fundamental parameters of the animation generation system by transitioning from a database-querying approach to a neural network generation approach. This parameter change transforms the system from one that retrieves pre-stored animations to one that generates animations computationally, fundamentally altering how animation variety is achieved and improving query performance.
3Productivity
If traditional motion matching technology is used, then animation generation capability is improved, but device complexity increases
Solution Approach 1:
The patent applies universality by designing a neural network that can handle multiple animation generation tasks through a single unified model. The network processes both trajectory features and bone features to generate various types of animations (running, jumping, squatting, etc.) using the same architecture, reducing system complexity compared to having separate specialized systems for different animation types.
Data Source
AI summary
This application describes an animation data generation method and apparatus, and a related product, which may be applied to scenarios such as a cloud technology, artificial intelligence (AI), intelligent transportation, assisted driving, digital human, virtual human, gaming, virtual reality, and extended reality (XR). Features of a virtual object in a virtual scene are obtained, and animation data of the virtual object is generated through a trained neural network based on the features. The use of the neural network omits storage of massive data into an internal memory and query of the massive data for a matched animation during generation of the animation data, and only requires storage of weight data related to the neural network in advance. Therefore, only a small internal memory is occupied, thereby avoiding problems such as large internal memory occupation and poor query performance during generation of the animation data.