Microscopic Video Annotation Support for Consistent AI Task Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in manual assembly of precision devices under a microscope, such as medical devices, is the high variability in work quality due to skill level differences among workers, and the vast amount of video data makes manual review impractical for ensuring task accuracy.
Innovation Solution
An annotation work support system that displays previously performed annotation information, including division positions and classifications, to assist in re-training AI models for improved task classification accuracy, using a stereoscopic microscope, control device, and monitor to facilitate efficient annotation work.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review of vast video data is performed to ensure task accuracy, then classification accuracy is improved, but work time and labor burden increase significantly
Solution Approach 1:
The system creates a copy of historical annotation information and classification criteria to assist current annotation work. By displaying previously performed annotations and the criteria used, the system enables annotators to quickly reference past decisions without re-reviewing entire video datasets, thus maintaining accuracy while reducing time investment.
Solution Approach 2:
The system provides feedback by displaying historical annotation information and classification criteria that were used in previous annotations. This feedback mechanism guides current annotators in making consistent decisions by showing what has been done before and why, reducing the need for manual review of vast video data while ensuring accuracy.
2Measurement precision
If re-training AI models is performed to improve task classification accuracy, then model performance is improved, but the burden and complexity of the process increases
Solution Approach 1:
The system performs preliminary action by storing and displaying historical annotation information and classification criteria before the re-training process begins. By having this information readily available during annotation work, the system simplifies the re-training process as annotators can efficiently update annotations based on displayed criteria rather than starting from scratch.
Solution Approach 2:
The system copies historical annotation information and classification criteria into the current working environment. This copying allows the re-training process to build upon existing annotated data and criteria rather than creating everything anew, reducing the burden and complexity of re-training while maintaining or improving model accuracy.
3Productivity
If annotation work is performed without reference to historical data, then work speed is improved, but consistency and reliability of annotations decrease
Solution Approach 1:
The system provides feedback by displaying historical annotation information and classification criteria alongside current video data. This feedback allows annotators to maintain high work speed while ensuring consistency, as they can quickly reference what has been annotated before and the criteria used, without slowing down to review entire historical datasets.
Solution Approach 2:
The system segments the annotation process by displaying only the relevant historical annotation information and classification criteria that pertain to the current task, rather than presenting the entire historical dataset. This segmentation allows annotators to access necessary reference information quickly while maintaining high productivity and annotation consistency.
Data Source
AI summary
A storage stores at least one trained models trained by using at least one microscopic videos, microscopic videos associated with each of the trained models, annotation information (including classifications for video segments included in the microscopic videos) indicating annotations assigned to the microscopic videos, and classification criterion information indicating a classification assignment criterion. The processor receives a selection of at least one trained model, acquires microscopic videos, annotation information, and classification criterion information associated with the selected trained model from the storage, displays the acquired microscopic videos and the acquired annotation information in association with each other on a display, and displays the acquired classification criterion information on the display.