Reproduce “AdaSin: Enhancing Hard Sample Metrics with Dual Adaptive Penalty for Face Recognition” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Qiqi Guo, Zhuowen Zheng, Guanghua Yang, Zhiquan Liu, Xiaofan Li, Jianqing Li, Jinyu Tian, Xueyuan Gong

Researchers from Jinan University introduce AdaSin, a novel loss function that dramatically improves face recognition accuracy by incorporating a dual adaptive penalty system based on sinusoidal difficulty metrics, achieving superior performance across multiple benchmarks while more effectively handling challenging samples through innovative curriculum learning techniques.

Key insights

  • Using sine of the angle between feature embeddings and class centers provides a more effective measure of sample difficulty
  • Applying adaptive penalties to both positive and negative similarities improves feature discrimination
  • Careful modulation of penalties through training stages helps balance easy and hard sample learning

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.