Training Device Model Selection via Feature Similarity Index
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing training models cannot quantitatively indicate their suitability for a specific training purpose, leading to inefficiencies in development as recognition performance tests are necessary for model selection.
Innovation Solution
A training device that acquires training images, calculates shared feature space features, compares these features with pre-stored existing features to determine similarity, and selects a base model for retraining based on this index, thereby improving development efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If recognition performance tests are conducted for each training model to determine suitability, then model selection accuracy is improved, but development time and computational resources are significantly consumed
Solution Approach 1:
The patent pre-calculates and stores feature vectors for multiple training models during the model training phase, before actual model selection is needed. This preliminary action allows the system to quickly compare new training images against pre-computed model features without performing full recognition performance tests, thereby reducing development time while maintaining selection accuracy
Solution Approach 2:
The patent replaces the traditional mechanical approach of conducting full recognition performance tests with a computational feature comparison method. By using pre-extracted feature vectors and similarity calculations, the system substitutes expensive, time-consuming performance evaluations with efficient mathematical operations, significantly reducing computational overhead and development time
2Measurement precision
If full recognition performance tests are performed for model selection, then model suitability is accurately determined, but computational resources and processing power are excessively consumed
Solution Approach 1:
The patent extracts essential feature vectors from training models and stores them separately, allowing the system to perform model selection based on feature comparison rather than executing full recognition performance tests. This extraction approach isolates the critical information needed for model selection while eliminating the need for computationally intensive full-model evaluations
Solution Approach 2:
The patent performs only the necessary partial action of feature vector comparison instead of complete recognition performance tests. By comparing pre-extracted features rather than running full model evaluations, the system achieves sufficient model suitability assessment with fraction of the computational resources required for complete performance testing
Data Source
AI summary
A training device includes: a training image acquiring unit that acquires a training image; a feature extracting unit that calculates a shared feature space feature of the training image; an existing feature acquiring unit that acquires a pre-stored trained model and an existing feature corresponding to the pre-stored trained model; a feature comparing unit that calculates a similarity between the shared feature space feature and the existing feature as an index; a model selecting unit that selects, as a base model, one of the trained models suitable for a purpose of training, on the basis of the index; a model training unit that performs retraining for the base model; a model evaluating unit that evaluates inference performance of the retrained base model; and a trained model outputting unit that outputs the retrained base model.