Class-Variant Margin Normalized Softmax Loss for Deep Face Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Deep face recognition models often struggle with class imbalance and softmax saturation during training.
• Existing softmax loss variations are insufficient to overcome these challenges in feature extraction.

Purpose of the Study:

• To propose a novel Class-Variant Margin (CVM) normalized softmax loss function.
• To address class imbalance and softmax saturation issues in deep face recognition.

Main Methods:

• Introduced a CVM normalized softmax loss incorporating true-class and false-class margins in the cosine space.
• The true-class margin targets class imbalance, while the false-class margin mitigates early softmax saturation.
• The proposed loss function has negligible computational overhead and is easily integrated into deep learning frameworks.

Main Results:

• Comprehensive experiments were conducted on LFW, YTF, and MegaFace datasets.
• The CVM loss demonstrated significant effectiveness in improving deep face recognition performance.
• The method successfully alleviates class imbalance and postpones softmax saturation.

Conclusions:

• The proposed CVM normalized softmax loss is an effective solution for deep face recognition.
• It enhances discriminative feature extraction by addressing key training challenges.
• The approach offers practical advantages due to its ease of implementation and low computational cost.