SynFace: Face Recognition with Synthetic Data

Qiu, Haibo; Yu, Baosheng; Gong, Dihong; Li, Zhifeng; Liu, Wei; Tao, Dacheng

doi:10.1109/iccv48922.2021.01070

Cited by 61 publications

(28 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although differential privacy (DP) [17,1] based methods have theoretical guarantees of privacy leakage and have been applied to easy datasets like MNIST [26], it is not applicable to generate large-scale high-resolution face dataset, because of low utility of the generated samples [53,9]. The recent works on generating privacy-preserving face dataset try to model the collect real faces with advanced GAN models and then synthesize fake face images with modifications [52,27,31,6,36]. For example, [36] propose to synthesize identity-ambiguous faces by mixing two label vectors.…”

Section: Face Dataset and Privacymentioning

confidence: 99%

“…The recent works on generating privacy-preserving face dataset try to model the collect real faces with advanced GAN models and then synthesize fake face images with modifications [52,27,31,6,36]. For example, [36] propose to synthesize identity-ambiguous faces by mixing two label vectors. Different from de-identification methods that aim to remove the identity of face images [52], we expect the privacy-preserving dataset can be used to train downstream identification models.…”

Section: Face Dataset and Privacymentioning

confidence: 99%

“…Nevertheless, collecting and releasing large-scale face datasets raise increasingly more concerns on the privacy leakage of identity membership [20,52] and attribute [31,6] of training samples in recent years. Generating large-scale privacy-preserving face datasets for downstream tasks is urgent and challenging for face recognition community [52,31,36]. This paper focuses on the membership privacy of a public face dataset instead of inferring specific training samples or model parameters [39,12].…”

Section: Introductionmentioning

confidence: 99%

“…These naive distortion methods reduce the privacy as well as semantics in a face image, thus producing unsatisfying recognition performance. The recent works try to generate privacy-preserving face dataset by synthesizing identity-ambiguous faces [36], de-identification faces [52,27] or attribute-removed faces [31,6] with adapted generative adversarial networks (GANs). They can generate as many synthetic faces as required for privacy-preserving face recognition task.…”

Section: Introductionmentioning

confidence: 99%

“…As shown in the central part of Fig. 1, there exists remarkable domain gap between generated and original faces, which leads to poor face recognition performance of GAN-based methods [36]. Another drawback of GAN based methods is that the generators trained on one dataset cannot be deployed on unseen identities.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

FaceMAE: Privacy-Preserving Face Recognition via Masked Autoencoders

Wang¹,

Zhao²,

Peng³

et al. 2022

Preprint

View full text Add to dashboard Cite

Face recognition, as one of the most successful applications in artificial intelligence, has been widely used in security, administration, advertising, and healthcare. However, the privacy issues of public face datasets have attracted increasing attention in recent years. Previous works simply mask most areas of faces or synthesize samples using generative models to construct privacy-preserving face datasets, which overlooks the trade-off between privacy protection and data utility. In this paper, we propose a novel framework FaceMAE, where the face privacy and recognition performance are considered simultaneously. Firstly, randomly masked face images are used to train the reconstruction module in FaceMAE. We tailor the instance relation matching (IRM) module to minimize the distribution gap between real faces and FaceMAE reconstructed ones. During the deployment phase, we use trained FaceMAE to reconstruct images from masked faces of unseen identities without extra training. The risk of privacy leakage is measured based on face retrieval between reconstructed and original datasets. Experiments prove that the identities of reconstructed images are difficult to be retrieved. We also perform sufficient privacy-preserving face recognition on several public face datasets (i.e.CASIA-WebFace and WebFace260M). Compared to previous state of the arts, FaceMAE consistently reduces at least 50% error rate on LFW, CFP-FP and AgeDB.

show abstract