Recognition Model Relearning via Timed Learning Image Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In automated driving systems, updating recognition models requires efficient relearning to maintain accuracy, but existing methods lack efficient mechanisms for selecting and utilizing learning images effectively.
Innovation Solution
An information processing apparatus and method that includes a collection timing control unit and a learning image collection unit to selectively gather and utilize learning images based on features and similarity to accumulated images, facilitating efficient relearning of recognition models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If learning images are collected continuously without timing control, then the quantity of learning data increases, but the relearning efficiency decreases due to redundant and similar images
Solution Approach 1:
The collection timing control unit performs preliminary action by determining optimal collection timings before actual learning image collection occurs. It controls when to collect learning images based on predicted needs, preventing redundant collection and ensuring timely acquisition of necessary training data for recognition model updates.
Solution Approach 2:
The system uses feedback from the recognition model's current performance and the accumulated learning image database to adjust collection timing. The collection timing control unit receives feedback about what types of images are needed and when collection should occur, creating a closed-loop system that improves relearning efficiency while maintaining appropriate data quantity.
2Adaptability or versatility
If all collected learning image candidates are used for relearning, then the diversity of training data increases, but the computational load and relearning time increase significantly
Solution Approach 1:
The learning image collection unit extracts only the necessary subset of learning images from all collected candidates. It takes out specific images that meet certain criteria (such as those most relevant to current recognition model needs or those providing maximum information gain) while discarding redundant ones, thus maintaining training diversity without incurring full computational costs of processing all collected images.
Solution Approach 2:
Instead of using all collected learning images (excessive action), the system applies partial action by selectively using only the portion of images that are most beneficial for relearning. This partial selection approach maintains adequate training data diversity while significantly reducing computational load and relearning time compared to processing the complete set of collected candidates.
3Ease of operation
If learning images are collected at fixed intervals, then the collection process is simple and systematic, but the recognition model cannot be updated efficiently when urgent updates are needed
Solution Approach 1:
The collection timing control unit implements dynamic collection timing that adapts to changing needs. Rather than fixed intervals, the system dynamically adjusts when to collect learning images based on current requirements, allowing both systematic regular collection and flexible urgent collection when recognition model updates are needed, thus combining simplicity with adaptability.
Solution Approach 2:
The collection timing control mechanism serves multiple functions: it performs systematic regular collection for routine model maintenance and also handles urgent on-demand collection when performance degradation is detected or new recognition scenarios emerge. This multi-functional approach maintains operational simplicity while providing the flexibility needed for various update scenarios.
Data Source
AI summary
The present technology relates to an information processing apparatus, an information processing method, and a program that enable efficient relearning of a recognition model. An information processing apparatus includes: a collection timing control unit configured to control a timing to collect a learning image candidate that is an image to be a candidate for a learning image to be used in relearning of a recognition model; and a learning image collection unit configured to select the learning image from among the learning image candidates that have been collected, on the basis of at least one of a feature of the learning image candidate or a similarity to the learning image that has been accumulated. The present technology can be applied to, for example, a system that controls automated driving.