Image Analysis Model Update Using Inference Result Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems fail to properly update detection models when changes in detection accuracy occur due to environmental or system changes, leading to incorrect detection of previously identified objects.
Innovation Solution
An image analysis system with a first inference unit using a trained model, a relearning unit for generating a retrained model, and a comparison unit to determine model updates based on detection accuracy comparisons between the two models, utilizing conditions such as improved detection rates and reduced false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a retrained model is used to improve detection accuracy, then detection accuracy is improved, but previously detected objects may not be detected correctly
Solution Approach 1:
The system performs preliminary comparison between the trained model and retrained model before deploying the retrained model. The comparison unit evaluates detection results from both models on the same input images, and only updates if the retrained model shows superior performance across multiple evaluation criteria, preventing premature deployment that could cause instability
Solution Approach 2:
The system implements a feedback mechanism where the comparison unit continuously evaluates detection results from both models and provides feedback on performance differences. This feedback loop ensures that model updates only occur when the retrained model demonstrably outperforms the trained model, maintaining reliability while improving accuracy
2Adaptability or versatility
If model retraining is performed to adapt to environmental changes, then adaptability is improved, but system complexity increases
Solution Approach 1:
The system segments the model update process into distinct functional units: a relearning unit that generates the retrained model, a comparison unit that evaluates both models, and an update determination unit that decides whether to deploy. This segmentation allows the complexity to be managed through modular design while maintaining adaptability
Solution Approach 2:
The comparison unit acts as an intermediary between the retrained model and the deployed model. It mediates the transition by evaluating detection results from both models and determining whether the retrained model should replace the trained model, thereby managing system complexity through a controlled intermediate evaluation step
Data Source
AI summary
An image analysis system and update method for a machine-learning model are to update a model in operation 111 to a retrained model 121, wherein in an image analysis server 1 which analyzes shot images, a first inference unit 11 performs inference using the trained model in operation 111, a relearning unit 14 performs relearning to generate a retrained model, a second inference unit 12 performs inference using a retrained model 121, and a comparison unit 15 compares the execution results of the first inference unit 11 and the second inference unit 12 to determine whether or not to update the model in operation 111 to the retrained model 121.