Graph-Based Pose and Motion Generation Across Skeleton Sizes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural IK and neural motion completion models are limited to generating poses and motions for entities with specific sizes and characteristics, requiring the creation of new datasets and models for each variation, which is time and resource-intensive.
Innovation Solution
A technique involving a graph representation of joints and proportions, combined with a neural network that generates poses and motions for different skeleton sizes and styles while preserving sparse constraints, using a single machine learning model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional neural IK models are trained on datasets specific to each skeleton size and entity characteristics, then the model can generate accurate poses for that specific entity type, but creating and training new models for each variation is time-consuming and resource-intensive
Solution Approach 1:
The patent applies universality by designing a single neural IK model that can handle multiple skeleton sizes and entity characteristics. The system uses a unified model architecture that processes input poses and generates output poses for different entity types without requiring separate trained models for each variation, thus reducing training time while maintaining pose generation accuracy across diverse entities.
Solution Approach 2:
The patent employs parameter changes by using proportional parameters that define skeleton characteristics. Instead of training separate models for different skeleton sizes, the system adjusts proportional parameters to represent different entity types and skeleton configurations, allowing a single model to adapt to various entity characteristics through parameter variation rather than retraining.
2Loss of time
If a single neural IK model is used for all skeleton sizes and entity characteristics, then time and resource overhead is reduced, but the model may struggle to maintain accuracy across diverse entity variations
Solution Approach 1:
The unified neural IK model is designed with universal input and output structures that can process poses for any entity type. The model maintains accuracy across diverse entities by using a consistent processing framework that adapts to different skeleton configurations through proportional parameters rather than requiring entity-specific model variants.
Solution Approach 2:
The system maintains pose generation accuracy across diverse entity variations by utilizing proportional parameters as adaptive inputs. These parameters encode skeleton size and characteristic information, allowing the single model to adjust its processing based on the specific entity type being posed, thereby preserving accuracy without requiring separate trained models for each entity variation.
3Adaptability or versatility
If conventional approaches create separate datasets for each skeleton size and entity type, then the model can learn entity-specific characteristics, but the overall system complexity and resource requirements increase significantly
Solution Approach 1:
The patent reduces system complexity by implementing a universal neural IK model that handles all entity types through a single unified architecture. Instead of maintaining multiple entity-specific models and datasets, the system uses one model that processes all entity variations, thereby reducing computational resources, storage requirements, and system complexity while maintaining the ability to generate entity-specific poses.
Solution Approach 2:
The system manages entity-specific characteristics through proportional parameters that encode skeleton size and type information. Rather than requiring separate datasets and models for each entity type, the approach uses parameter variation to represent different entity characteristics, simplifying the system architecture while preserving adaptability to diverse entity types through parameter-based differentiation.
Data Source
AI summary
One embodiment of the present invention sets forth a technique for generating a pose for a virtual character. The technique includes determining a graph representation of one or more sets of joints in the virtual character based on (i) constraints associated with one or more joints included in the set(s) of joints and (ii) proportions associated with pairs of joints included in the set(s) of joints. The technique also includes generating, via execution of a neural network, a set of updated node states for the set(s) of joints based on the graph representation. The technique further includes generating, based on the updated node states, one or more output poses that correspond to the set(s) of joints, wherein the output pose(s) include (i) a first set of joint positions for the set(s) of joints, (ii) a first set of joint orientations for the set(s) of joints, and (iii) the proportions.