CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition

By Yuge Huang, Yuhan Wang, Ying Tai, Xiaoming Liu, Pengcheng Shen, Shaoxin Li, Jilin Li, Feiyue Huang. 2020

660 citations recorded by the source corpus.

As an emerging topic in face recognition, designing margin-based loss functions can increase the feature margin between different classes for enhanced discriminability. More recently, the idea of mining-based strategies is adopted to emphasize the misclassified samples, achieving promising results. However, during the entire training process, the prior methods either do not explicitly emphasize the sample based on its importance that renders the hard samples not fully exploited; or explicitly emphasize the effects of semi-hard/hard samples even at the early training stage that may lead to convergence issue. In this work, we propose a novel Adaptive Curriculum Learning loss (CurricularFace) that embeds the idea of curriculum learning into the loss function to achieve a novel training strategy for deep face recognition, which mainly addresses easy samples in the early training stage and hard ones in the later stage. Specifically, our CurricularFace adaptively adjusts the relative importance of easy and hard samples during different training stages. In each stage, different samples are assigned with different importance according to their corresponding difficultness. Extensive experimental results on popular benchmarks demonstrate the superiority of our CurricularFace over the state-of-the-art competitors.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation (2022, methodology): “We implemented some methods in the face recognition community that might be relevant to our work, including (1) Adaptive-margin-based methods (CurricularFace [17], AdaCos [41], ElasticFace [3], AMRLoss [42]), such as adjusting the margin value according to the intra/inter-class angles.”
Encoding in Style: a StyleGAN Encoder for Image-to-Image Translation (2020, methodology): “To make sure the similarity score is independent of our loss function, we utilize the CurricularFace [15] method for evaluation.”
HyperInverter: Improving StyleGAN Inversion via Hypernetwork (2021, methodology): “For the human facial domain, we measure the ability to preserve the subject identity of each inversion method by computing the identity similarity between the source images and the reconstructed ones using an off-the-shelf face recognition model CurricularFace [19].”
E2Style: Improve the Efficiency and Effectiveness of StyleGAN Inversion (2021, methodology): “The method we used here to calculate the identity similarity is Curricularface [60], instead of ArcFace [52] which we used for training.”
WebFace260M: A Benchmark Unveiling the Power of Million-Scale Deep Face Recognition (2021, background): “CosFace [68], ArcFace [12] and CurricularFace [27]) in one Gluon codebase with the hyper-parameters referred to the original papers.”
SphereFace Revived: Unifying Hyperspherical Face Recognition (2021, background): “Instead of a static characteristic function, designing a dynamic characteristic function could be beneficial [37], [40].”
HyperStyle: StyleGAN Inversion with HyperNetworks for Real Image Editing (2021, methodology): “We additionally measure identity similarity using a pre-trained facial recognition network [27].”
A Simple Baseline for StyleGAN Inversion (2021, methodology): “The method we used here to calculate the identity similarity is Curricularface [21], instead of ArcFace [11] which we used for training.”
AdaFace: Quality Adaptive Margin for Face Recognition (2022, methodology): “Note that the augmentation hurts the performance of CurricularFace, which is in line with our assumption that augmentation is a tradeoff between a positive effect from getting more data and a negative effect from unidentifiable images.”
ElasticFace: Elastic Margin Loss for Deep Face Recognition (2021, methodology): “In CALFW, ElasticFace-Arc scored a close third with 96.17% accuracy, with the lead going to the CurricularFace [10] (CVPR2020) with 96.20% accuracy.”
Third Time's the Charm? Image and Video Editing with StyleGAN3 (2022, methodology): “In addition to the inference time required by each inversion technique, we report the L 2 distance, the LPIPS [79] distance, identity similarity [28], and MS-SSIM [72] score between the reconstructions and their sources.”
Contrastive Learning for Diverse Disentangled Foreground Generation (2022, methodology): “For face, we use CurricularFace [16], and for bird and car, we train a new classifier using the VGG architecture [38].”
Adaptive Curriculum Learning (2021, background): “Similarly, [14] proposed Cur-ricularFace, a dynamic curriculum learning for deep face recognition by adding a modulation coefﬁcient function to the softmax-based loss function.”