Killing Two Birds with One Stone:Efficient and Robust Training of Face Recognition CNNs by Partial FC

TL;DR

This paper introduces Partial FC, a sparse update method for face recognition CNNs that reduces memory, computation, and class conflict issues by selecting a subset of class centers during training, improving efficiency and robustness.

Contribution

The paper proposes Partial FC, a novel sparse updating strategy for the fully connected layer in face recognition CNNs, addressing scalability and data imbalance challenges.

Findings

Significant reduction in memory and computation costs.

Improved robustness against inter-class conflict and long-tailed data.

Effective across various datasets and backbone architectures.

Abstract

Learning discriminative deep feature embeddings by using million-scale in-the-wild datasets and margin-based softmax loss is the current state-of-the-art approach for face recognition. However, the memory and computing cost of the Fully Connected (FC) layer linearly scales up to the number of identities in the training set. Besides, the large-scale training data inevitably suffers from inter-class conflict and long-tailed distribution. In this paper, we propose a sparsely updating variant of the FC layer, named Partial FC (PFC). In each iteration, positive class centers and a random subset of negative class centers are selected to compute the margin-based softmax loss. All class centers are still maintained throughout the whole training process, but only a subset is selected and updated in each iteration. Therefore, the computing requirement, the probability of inter-class conflict, and the frequency of passive update on tail class centers, are dramatically reduced. Extensive experiments across different training data and backbones (e.g. CNN and ViT) confirm the effectiveness, robustness and efficiency of the proposed PFC. The source code is available at \https://github.com/deepinsight/insightface/tree/master/recognition.