Partial FC method enables training face recognition models with millions of identities

Researchers have developed a novel method called Partial FC (PFC) to efficiently train face recognition models with millions of identities on a single machine. This technique approximates the full softmax classifier by activating only a sampled subset of negative class centers per mini-batch, while still preserving all positive class centers. PFC significantly reduces memory, computation, and communication overhead, enabling scalable training to tens of millions of classes with competitive accuracy and improved efficiency. The method also demonstrates robustness to label noise and long-tailed identity distributions. AI

IMPACT This method could significantly reduce the computational cost and hardware requirements for training large-scale face recognition systems.

RANK_REASON The cluster contains an academic paper detailing a new method for training machine learning models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• ICCV21-MFR
• IJB-C
• Partial FC
• Softmax

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →