Identification Information Assignment Apparatus for Automated Training Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual annotation of identification information for machine learning data is labor-intensive and time-consuming, especially with large amounts of data, which hinders the efficient generation of high-quality learning data.
Innovation Solution
An identification information assignment apparatus that acquires image data, uses a learning model to automatically assign identification information, and updates the model based on the assigned data, allowing for efficient generation of learning data by selecting and annotating image data in stages with improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation is used to assign identification information to image data, then accuracy of annotation can be ensured, but labor time and processing time increase significantly
Solution Approach 1:
The system performs preliminary annotation using a learning model before manual review, pre-processing the image data to assign identification information. This preliminary action reduces the subsequent manual work required while maintaining accuracy, as annotators only need to review and correct the pre-annotated data rather than annotate from scratch.
Solution Approach 2:
A learning model serves as an intermediary between the image data and the final annotated result. The model generates initial annotation predictions that bridge the gap between raw data and manually verified annotations, reducing the direct manual effort required while preserving accuracy through subsequent verification.
2Quantity of substance
If a large amount of image data is processed manually, then comprehensive learning data can be generated, but productivity decreases due to labor intensity
Solution Approach 1:
The annotation process is segmented into two distinct phases: automated annotation by the learning model and manual verification/correction. This segmentation allows the system to process large volumes of data through the efficient automated phase while maintaining quality through selective manual intervention, thereby increasing overall productivity without sacrificing data comprehensiveness.
Solution Approach 2:
The system changes the parameter of annotation methodology from purely manual to a hybrid automated-manual approach. By adjusting the degree of automation and utilizing the learning model's capabilities, the system can process larger quantities of data at higher speeds while maintaining acceptable accuracy levels through subsequent verification.
3Productivity
If automated annotation using a learning model is used, then processing speed increases, but annotation accuracy may decrease
Solution Approach 1:
The system implements a feedback mechanism where manually verified and corrected annotations are used to retrain and update the learning model. This continuous feedback loop improves the model's accuracy over time, allowing it to produce more reliable initial annotations that require less manual correction, thereby maintaining both speed and accuracy.
Solution Approach 2:
The learning model acts as an intermediary that generates initial annotations which are then refined through manual verification. This intermediary role allows the system to leverage the speed of automated processing while using human expertise to correct errors, achieving both high processing speed and maintained accuracy.
4Measurement precision
If iterative model updating is performed, then annotation accuracy improves, but system complexity increases
Solution Approach 1:
The system performs model updating as a preliminary or periodic action rather than continuously for every annotation task. By pre-updating the model with verified data and then using it for batch processing, the system achieves improved accuracy without the complexity of real-time iterative updates during annotation operations.
Data Source
AI summary
It reduces labor and time to generate training data for the training model. An identification information assignment apparatus includes an acquirer configured to acquire a plurality of pieces of image data, an assigner configured to assign identification information to image data selected from the plurality of pieces of image data by using a learning model after learning, and an updater configured to update the learned model using the image data to which the identification information is assigned, wherein the assigner assigns identification information to a rest of the image data acquired by the acquirer using the learned model that has been updated.