Boosting Unconstrained Face Recognition with Targeted Style Adversary
This addresses the domain generalization issue in face recognition for applications like surveillance or authentication, but it is incremental as it builds on existing feature-based augmentation approaches.
The paper tackles the problem of deep face recognition models struggling with inputs from domains beyond their training data by introducing a method to expand training data through feature statistic interpolation, resulting in outperforming or matching competitors with a 70% improvement in training speed and 40% less memory consumption.
While deep face recognition models have demonstrated remarkable performance, they often struggle on the inputs from domains beyond their training data. Recent attempts aim to expand the training set by relying on computationally expensive and inherently challenging image-space augmentation of image generation modules. In an orthogonal direction, we present a simple yet effective method to expand the training data by interpolating between instance-level feature statistics across labeled and unlabeled sets. Our method, dubbed Targeted Style Adversary (TSA), is motivated by two observations: (i) the input domain is reflected in feature statistics, and (ii) face recognition model performance is influenced by style information. Shifting towards an unlabeled style implicitly synthesizes challenging training instances. We devise a recognizability metric to constraint our framework to preserve the inherent identity-related information of labeled instances. The efficacy of our method is demonstrated through evaluations on unconstrained benchmarks, outperforming or being on par with its competitors while offering nearly a 70\% improvement in training speed and 40\% less memory consumption.