Feature Conversion Model for Object Recognition Compatibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition systems face challenges when updating extraction models, as features extracted using higher version models may not match with those in the matching library, requiring re-extraction of features from a large number of samples, leading to significant time consumption.
Innovation Solution
The method involves converting features extracted using a first extraction model to the feature space of a second extraction model using a feature conversion model, allowing for matching with features in the matching library without the need for re-extraction, thereby reducing processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a higher version extraction model is used to extract features, then recognition accuracy is improved, but the features cannot be matched with the matching library formed by a lower version extraction model
Solution Approach 1:
A feature conversion model is introduced as an intermediary component between the higher version extraction model and the matching library. This conversion model transforms features extracted by the higher version model into a format compatible with the lower version model's feature space, enabling successful matching without sacrificing recognition accuracy.
Solution Approach 2:
The feature conversion model changes the parameters or representation of extracted features to match the feature space of the lower version extraction model. By transforming the feature parameters rather than re-extracting them, the system maintains compatibility with the existing matching library while using the more accurate higher version model.
2Adaptability or versatility
If re-extraction of features from all samples is performed to form a new matching library, then feature compatibility is improved, but time consumption increases significantly
Solution Approach 1:
The feature conversion model is trained in advance on a subset of samples to learn the mapping between higher version and lower version feature spaces. This preliminary training enables the system to perform rapid feature conversion during actual recognition tasks, avoiding the need for time-consuming re-extraction of all matching library features.
Solution Approach 2:
Instead of re-extracting features from all samples in the matching library, the system extracts features only from a target subset of samples using the higher version model, then applies the pre-trained conversion model to transform these features. This selective extraction combined with conversion significantly reduces processing time while maintaining compatibility.
3Productivity
If a feature conversion model is trained on a target subset of samples, then training efficiency is improved, but the subset selection process becomes more complex
Solution Approach 1:
The system trains the feature conversion model on a partial subset of samples rather than the entire dataset. This partial training approach achieves sufficient conversion capability with reduced computational resources and time, while the subset selection criteria are designed to be simple enough to avoid excessive complexity.
Data Source
AI summary
Methods and apparatuses for performing object recognition are provided. The method includes: by converting a first feature of a target object extracted using a first extraction model to a feature space of a second extraction model through a first feature conversion model, obtaining a second feature of the target object in the feature space; by matching the second feature of the target object with features of objects in a matching library, obtaining a target feature matched with the second feature of the target object; determining an object to which the target feature belongs as a matching object of the target object. Thus, the present disclosure can save time length of object recognition.