Lifelog Camera Training Data Segmentation for Balanced Composition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Lifelog cameras that automatically capture images without user instruction often capture unwanted moments due to the need for large amounts of training data, which can lead to a disproportionate emphasis on user preferences, making it difficult to perform automatic image capturing and processing with generally preferable compositions, thereby deteriorating user-friendliness.
Innovation Solution
An information processing apparatus with a control circuit that accesses memory storing training data to perform learning, using both first and second training data, where the second training data is updated with characteristic data from images without updating the first training data, to balance user preferences and maintain image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of training data is used to learn user preferences, then the learning accuracy is improved, but the learning result places disproportionate emphasis on user preferences, making it difficult to perform automatic image capturing with generally preferable compositions
Solution Approach 1:
The training data is segmented into two distinct sets: first training data containing a large amount of data for learning user preferences, and second training data containing a limited amount of data for maintaining balanced composition. This segmentation allows the system to simultaneously capture user preferences and maintain general image quality without being overly influenced by individual user choices.
Solution Approach 2:
Different training data sets are assigned different qualities and roles: the first training data is used for learning user preference patterns, while the second training data with upper limit on characteristic data pieces is used for maintaining balanced and generally preferable compositions. This local differentiation of data quality and role resolves the contradiction between capturing user preferences and maintaining general image quality.
2Productivity
If the lifelog camera periodically performs automatic image capturing, then continuous recording is achieved, but video images that are not desired by the user are acquired and video images of desired moments cannot be captured
Solution Approach 1:
The system uses learning processing to analyze user preferences from training data and provides feedback in the form of predicted user preferences for automatic image capturing. This feedback mechanism allows the camera to capture images that align with user preferences rather than merely recording continuously regardless of user intent.
Solution Approach 2:
The system changes the parameter of image capturing from fixed periodic timing to preference-based timing by using learning results. The learning circuit processes training data to generate preference parameters that dynamically guide when images should be captured, replacing static periodic capturing with adaptive preference-driven capturing.
3Measurement precision
If the second training data is updated with characteristic data from images, then user preferences are accurately reflected, but the number of pieces of characteristic data is limited to maintain balanced composition
Solution Approach 1:
Instead of updating training data with all characteristic data from captured images (excessive action), the system updates only with a limited number of characteristic data pieces (partial action). This partial update approach, enforced by the upper limit on second training data, prevents overfitting to user preferences while still capturing essential preference patterns.
Data Source
AI summary
An information processing apparatus is provided which divides data for learning into first training data and second training data, stores the first training data and the second training data into a memory, and updates only the second training data without updating the first training data in a case where new characteristic data is obtained.