3D Asset Shape Synthesis Without Manual Part Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Producing 3D assets for computer simulations such as computer games is time-intensive, and even with machine learning, manual labeling of part segments is required, which is time-consuming.
Innovation Solution
A machine learning pipeline automatically generates 3D assets without explicit meta data by segmenting assets into parts, calculating hierarchical placement and similarity metrics, and ranking variations using shape synthesis and energy metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual labeling of part segments is used to generate 3D assets, then the quality and control of asset generation is improved, but the time required for asset generation increases significantly
Solution Approach 1:
The system performs automatic segmentation and part identification without requiring manual human intervention. The machine learning model processes 3D assets autonomously, extracting part segments, generating adjacency graphs, and creating shape descriptors automatically, thereby eliminating the time-consuming manual labeling process while maintaining generation quality
Solution Approach 2:
The patent replaces the manual mechanical process of human labeling with an automated machine learning-based system. The ML model substitutes human operators by automatically segmenting assets, calculating hierarchical relationships, and generating shape variations, thus reducing time consumption while preserving the precision needed for quality asset generation
2Extent of automation
If machine learning is used to aid in asset generation, then the automation level is improved, but manual meta data labeling is still required which reduces efficiency
Solution Approach 1:
The machine learning pipeline is designed to be fully autonomous, automatically performing segmentation, part identification, adjacency graph generation, and shape variation creation without requiring any manual meta data input. The system serves itself by processing raw 3D assets directly and generating outputs end-to-end, maximizing both automation level and productivity
Solution Approach 2:
The system performs all necessary preprocessing steps automatically as part of the ML pipeline, including segmentation and feature extraction, before generating shape variations. This eliminates the need for separate manual meta data preparation steps, thereby improving overall efficiency while maintaining high automation
3Productivity
If shape variations are generated automatically, then the productivity is improved, but the quality control and plausibility of generated shapes may deteriorate
Solution Approach 1:
The system employs a ranking metric that evaluates generated shape variations based on shape energy metrics and part similarity. This feedback mechanism automatically assesses the plausibility of each generated variation and ranks them, ensuring that only high-quality, plausible shapes are selected while maintaining high productivity through automated evaluation
Solution Approach 2:
The patent replaces manual quality review with an automated ranking system based on shape energy metrics. The ML model automatically evaluates and ranks generated variations, substituting human quality control with an efficient automated system that maintains precision while enabling high-speed generation
Data Source
AI summary
A machine learning (ML) pipeline works solely on 3D assets without any meta data such as part labels or part structure or part similarity metrics given explicitly alongside the 3D assets. The time required to generate assets is reduced by splitting the process into an online and offline stage. A ML retrieval model based on text and/or image input selects a template asset and candidate parts for generation. A ranking metric is formulated after generating variations from a shape synthesis module using part similarity metrics to rank generated assets. The ranking score closely matches the human perception. The metric can also be used to weed out defective assets generated without human intervention.