Face Verification Evaluation System Using Segmented Pair Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The evaluation of face verification, especially with large datasets, requires significant computational resources and storage, leading to low efficiency and high costs.
Innovation Solution
An evaluating method and system for face verification that involves obtaining querying and training datasets, generating verification pairs and an evaluation list, determining feature information, establishing correspondences between identifiers and features, and determining evaluation results to improve efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large-scale dataset is used to improve face verification security performance, then the accuracy and reliability of face verification are improved, but the computational cost and storage cost increase significantly
Solution Approach 1:
The patent segments the large-scale evaluation process into multiple stages: first generating verification pairs from the dataset, then evaluating these pairs through feature extraction and comparison. This segmentation allows the system to handle large datasets by processing them in manageable chunks rather than requiring simultaneous computation of all pairs, thereby reducing peak computational cost while maintaining evaluation thoroughness.
Solution Approach 2:
The patent performs preliminary actions by pre-generating verification pairs and organizing them into evaluation lists before the actual face verification computation. This preliminary organization of data structures and identification of verification pairs enables the main evaluation process to proceed more efficiently, reducing the computational burden during the actual verification phase.
2Reliability
If a large-scale dataset is used to improve face verification security performance, then the accuracy and reliability of face verification are improved, but the storage cost increases significantly
Solution Approach 1:
The patent extracts only the essential elements needed for verification from the large dataset - specifically, verification pairs consisting of query images and gallery images with their corresponding identifiers. By extracting and storing only these critical verification relationships rather than the entire dataset, the system maintains the ability to perform secure face verification while significantly reducing storage requirements.
3Measurement precision
If intensive computation is performed to evaluate large-scale face verification data, then the accuracy of evaluation is improved, but the evaluation efficiency decreases
Solution Approach 1:
The patent segments the evaluation process into distinct phases: verification pair generation, feature extraction, and result determination. Each phase processes data in a structured manner, allowing accurate computation to be performed on smaller, segmented portions of the data rather than attempting intensive computation on the entire dataset simultaneously, thereby maintaining accuracy while improving overall evaluation efficiency.
Solution Approach 2:
The patent performs preliminary organization of verification pairs and evaluation lists before the main evaluation computation. This preliminary structuring of data enables the subsequent accurate evaluation to proceed more efficiently by avoiding redundant computations and data organization steps during the main evaluation phase.
Data Source
AI summary
An evaluating method and a system for face verification, and a computer storage medium. The method comprises: obtaining a querying dataset and a training dataset, wherein the querying dataset comprises querying images, and the training dataset comprises training images; generating verification pairs and an evaluation list respectively according to the querying dataset and the training dataset; wherein the evaluation list comprises the querying images and the training images; determining feature information based on the evaluation list; establishing correspondences between identifiers and features according to first identifiers of the querying images, second identifiers of the training image, and the feature information; and determining evaluation results of the verification pairs according to the correspondences between the identifiers and the features.