Face recognition model training method and device for difficult example samples
By constructing a biased centrality calculation and centrality prediction network for the face recognition model and updating the loss function, the problem of the lack of mining of difficult examples in the loss function of the existing technology is solved, and the accuracy of the model is improved.
CN115937929BActive Publication Date: 2026-06-16SHENZHEN XUMI YUNTU SPACE TECH CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN XUMI YUNTU SPACE TECH CO LTD
- Filing Date
- 2022-10-26
- Publication Date
- 2026-06-16
AI Technical Summary
⚠Technical Problem
The loss function of existing face recognition models lacks effective mining of difficult examples, resulting in low accuracy of the final trained model.
⚗Method used
The face recognition model is trained for the first time by acquiring a face training set, the skewed centrality of the samples is calculated, and a centrality prediction network parallel to the fourth-stage network is constructed. The loss function is updated using the predicted centrality, and then a second training is performed.
🎯Benefits of technology
It improves the ability of face recognition models to mine difficult examples and enhances the accuracy of the models.
✦ Generated by Eureka AI based on patent content.
Abstract
The present disclosure relates to the technical field of face recognition, and provides a face recognition model training method and device for difficult example samples. The method comprises: obtaining a face training set, and performing first training on a face recognition model using the face training set; calculating a biased central degree corresponding to each sample in the face training set using the face recognition model after the first training; constructing a central degree network in parallel with the fourth stage network of the face recognition model at the fourth stage network of the face recognition model; determining a predicted central degree corresponding to each sample using the face recognition model after the construction of the central degree network; updating a loss function of the face recognition model according to the biased central degree and the predicted central degree corresponding to each sample; and performing second training on the face recognition model after the updating of the loss function using the face training set. The above technical means solves the problem that the loss function of the face recognition model lacks effective mining of difficult example samples in the prior art.
Need to check novelty before this filing date? Find Prior Art