Feature Quantity Data Division for Domain Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing domain adaptation technologies face challenges in accurately classifying images from a target domain due to insufficient training of feature quantity extraction and domain identification sections, and they do not effectively adapt to the environment of each user using a machine learning model trained by a CG image.
Innovation Solution
A learning apparatus with a classification learning section, a dividing section, and a domain identification learning section that trains the feature quantity extraction and classification sections by dividing feature quantity data into partial data, allowing for more accurate domain identification and classification of images into multiple classes, suitable for various user environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the entire feature quantity data is input to the domain identification section, then the training process is simple, but the classification accuracy of target domain images is insufficient
Solution Approach 1:
The patent divides the entire feature quantity data into multiple pieces of partial feature quantity data corresponding to different classes. Each partial feature quantity data is then input to the domain identification section for separate training. This segmentation approach increases classification accuracy by allowing the model to learn domain-specific features for each class independently, while maintaining manageable training complexity through modular processing.
2Ease of manufacture
If a machine learning model trained by CG images is used, then the initial classification capability is provided, but the model does not adapt to diverse user environments
Solution Approach 1:
The patent uses CG images to pre-train the classification model before deployment, providing initial classification capability. Then, during actual use, the model performs domain identification learning by comparing features from real captured images with those from CG images, adapting to the specific user environment. This preliminary action approach balances quick deployment with environmental adaptability.
3Ease of operation
If domain adaptation is performed using traditional methods, then the process is straightforward, but the training of feature quantity extraction and domain identification sections is not sufficiently accurate
Solution Approach 1:
The patent implements a feedback mechanism where the domain identification section compares features from captured images with those from CG images, and the classification learning section uses this comparison result to adjust the feature quantity extraction section. This feedback loop continuously improves training accuracy by leveraging the known ground truth of CG image domains while maintaining operational simplicity through automated iterative optimization.
Data Source
AI summary
A classification learning section executes training of a feature quantity extraction section and training of a classification section resulting from a comparison between an output generated when feature quantity data is inputted to the classification section and training data regarding a plurality of classes associated with a source domain training image. A dividing section divides feature quantity data outputted from the feature quantity extraction section in accordance with input of an image into a plurality of pieces of partial feature quantity data corresponding to the image including a feature map of one or more of the classes. A domain identification learning section executes training of the feature quantity extraction section resulting from a comparison between an output generated when partial feature quantity data corresponding to the image is inputted to a domain identification section and data indicating whether the image belongs to a source domain or to the target domain.