Synthetic Training Data Generation With Auto Annotation for Vision Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The current methods for generating training data for machine learning computer vision models are time-consuming and require significant manual effort, expertise in computer vision, and specialized knowledge, making it difficult for non-experts to use artificial intelligence effectively.
Innovation Solution
A computer-implemented method and system for automatically generating synthetic training data sets by processing user-defined 2D or 3D models of objects of interest, determining render parameters, and producing annotated training images, which can be used to train machine learning computer vision models with minimal human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual data collection and labelling methods are used, then training data can be obtained, but the process is time-consuming and requires significant manual effort and expertise
Solution Approach 1:
The patent uses synthetic data generation by rendering 3D models to create training images, copying the essential visual characteristics of real objects without requiring physical photographs. This eliminates manual data collection while maintaining data quality through controlled rendering parameters.
Solution Approach 2:
The system automatically generates annotations and training data structures through the rendering process itself, without requiring manual labelling. The rendering engine self-generates ground truth information such as object positions, orientations, and segmentation masks directly from the 3D model data.
2Measurement precision
If manual data labelling is performed, then annotated training data is produced, but significant expertise in computer vision is required
Solution Approach 1:
The rendering system automatically generates precise annotations including bounding boxes, segmentation masks, and object attributes directly from the 3D models during the image generation process. This eliminates the need for manual labelling by experts while maintaining high annotation accuracy through programmatic precision.
Solution Approach 2:
The patent introduces a rendering engine as an intermediary between 3D models and training data requirements. This intermediary automatically translates geometric model data into annotated image formats, bridging the gap between simple model input and complex annotated output without requiring user expertise in either domain.
3Measurement precision
If specialized knowledge is required for data generation, then high-quality training data can be created, but the process becomes difficult for non-experts to use
Solution Approach 1:
The system copies verified high-quality data generation methodologies from professional pipelines into an automated rendering system that can be accessed through simple interfaces. Users benefit from expert-level data quality without needing expert knowledge, as the rendering engine encapsulates complex algorithms internally.
Data Source
AI summary
A computer-implemented method for generating a synthetic training data set for training a machine learning computer vision model for performing at least one user defined computer vision task, in which spatially resolved sensor data are processed and evaluated with respect to at least one user defined object of interest, including receiving at least one model of a user defined object of interest; determining at least one render parameter and multiple render parameters; generating a set of training images by rendering the at least one model of the object of interest based on the at least one render parameter; generating annotation data for the set of training images with respect to the at least one object of interest; and providing a training data set including the set of training images and the annotation data for being output to the user and/or for training the computer vision model.