Recognition Model Selection by Attribute Matching and Data Diversity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques struggle to select a model with high recognition performance that matches the attribute of the recognition target data from a plurality of models, especially when the models are learned using different data sets.
Innovation Solution
An information processing apparatus that acquires information about learning data and recognition target data attributes to select a model based on the matching degree and diversity of attributes, ensuring the selected model includes the target data attributes and minimizes extra learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a model learned by using a learning data set formed by various kinds of data is used, then the model can cope with various kinds of recognition target data, but the recognition performance for specific attribute data is lower compared to models learned with limited data
Solution Approach 1:
The patent segments the learning data sets into multiple categories based on data attributes (e.g., imaging conditions, object types). Instead of using a single diverse data set, the system divides and conquers by creating specialized models for different data segments, allowing each model to achieve high performance on its specific segment while collectively covering various recognition scenarios
Solution Approach 2:
The patent introduces a dynamic model selection mechanism that adapts the choice of learning data set based on the attributes of the recognition target data. The system dynamically determines which pre-learned model to use by comparing target attributes with training attributes, ensuring the most appropriate model is selected for each specific recognition task
2Measurement precision
If a model learned by using a learning data set formed by limited data is used, then the recognition performance for specific attribute data is higher, but the model can only cope with little variations of recognition target data
Solution Approach 1:
The patent creates a universal model selection system that manages multiple specialized models. Each model is trained on limited data for high performance on specific attributes, but the overall system achieves universality by selecting the appropriate specialized model based on the recognition target's attributes, making the system capable of handling various kinds of data effectively
3Ease of operation
If only models with similar imaging conditions are selected, then the selection process is simplified, but it becomes difficult to select a model learned by using data corresponding to the attribute of a recognition target
Solution Approach 1:
The patent changes the selection parameters from simple imaging condition similarity to comprehensive attribute matching. The system evaluates multiple attributes (imaging conditions, object types, environmental factors) and selects models based on the degree of matching between target attributes and training attributes, achieving both ease of operation through automated comparison and high precision through multi-dimensional attribute evaluation
Data Source
AI summary
An information processing apparatus comprising: a first acquiring unit configured to acquire information about learning data used in learning of each of a plurality of prelearned models for recognizing input data; a second acquiring unit configured to acquire information indicating an attribute of recognition target data; and a model selecting unit configured to select a model to be used in recognition of the recognition target data from the plurality of models, based on a degree of matching between the attribute of the recognition target data and an attribute of the learning data used in learning of each of the plurality of models, and on diversity of the attribute of the learning data used in learning of each of the plurality of models.