Angle-Converted Domain Adaptation for Image Classification Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Domain adaptation techniques face inefficiencies when training images have limited variation in shooting angles.
Innovation Solution
A training apparatus and method that includes feature extraction, angle conversion, and class prediction processes, utilizing source and target domain data to generate and update feature and class predictors, with losses computed to enhance training efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If domain adaptation techniques are used with limited shooting angle variation training images, then training efficiency is improved, but classification accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by performing angle conversion on source domain feature values before classification. The angle converter generates converted feature values that simulate different shooting angles, preparing the data in advance to compensate for limited angle variation in training images, thereby improving both training efficiency and classification accuracy
Solution Approach 2:
The patent introduces an angle converter as an intermediary component between feature extraction and classification. This intermediary transforms source domain feature values into converted feature values with simulated angle variations, bridging the gap between limited training data and diverse real-world classification scenarios
2Ease of operation
If only target domain data is used for training, then training simplicity is improved, but model performance deteriorates due to limited labeled data
Solution Approach 1:
The patent merges source domain and target domain data through domain adaptation. It extracts feature values from both domains, converts source domain features to match target domain characteristics, and jointly trains the classification model, combining the simplicity of target domain training with the performance benefits of source domain knowledge
Solution Approach 2:
The patent applies parameter changes by transforming source domain feature values into converted feature values that match the distribution and characteristics of target domain features. This parameter transformation enables effective utilization of source domain data while maintaining compatibility with target domain classification requirements
Data Source
AI summary
To provide an efficient training process even in a case where training images having a limited variation of shooting angles are available.Solution to ProblemA training apparatus (10) comprises: feature extraction section (11) for extracting source domain feature values from input source domain image data and for extracting target domain feature values from input target domain image data; angle conversion section (12) for generating converted source domain feature values by converting the source domain feature values as if the converted source domain feature values are extracted from source domain image data having different angles from the input source domain image data, and generating converted target domain feature values by converting the target domain feature values as if the converted target domain feature values are extracted from target domain image data having different angles from the input target domain image data; class prediction section(13) for predicting source domain class prediction values from the source domain feature values and the converted source domain feature values, and predicting target domain class prediction values from the target domain feature values and the converted target domain feature values; and updating section (14) for updating at least one of (i) the feature extraction section, (ii) the angle conversion section, and (iii) the class prediction section.