A data-driven approach to cleaning large face datasets

By Hongwei Ng, Stefan Winkler. 2014

723 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age (2019, methodology): “We used the facial embedding based on ResNet-34 from dlib, which was trained from the FaceScrub dataset [40], the VGG-Face dataset [41] and other online sources, which are likely dominated by the White faces.”
Unsupervised Cross-Domain Image Generation (2016, methodology): “We evaluate the visual quality that is obtained per person and not just per image, by testing DTN on the Facescrub dataset (Ng & Winkler, 2014).”
Exploiting Unintended Feature Leakage in Collaborative Learning (2018, methodology): “FaceScrub [40] contains 76,541 50x50 RGB images for 530 individuals with the gender label: 52.5% are labeled as male, the rest as female.”
Machine Learning Models that Remember Too Much (2017, methodology): “task, we use additional attribute labels [42]. Each image is rescaled to 67x42 RGB pixels from its original size, so that all images have the same size. FaceScrub is a dataset of URLs for 100K images [59]. The tasks are face recognition and gender classification. Some URLs have expired, but we were ”
Defeating Image Obfuscation with Deep Learning (2016, methodology): “On the FaceScrub [32] dataset of photos of over 530 celebrities, we achieve accuracy of 57% against mosaicing with 16 × 16 windows and 40% against P3 with threshold 20, vs. 0.19% for random guessing.”
Diversity in Faces (2019, methodology): “The gender annotations used two class labels (male and female) and the age group labeling used seven classes ([0-3],[4-12],[13-19],[20-30],[31-45],[46-60],[61-]), as well as a continuous age value to be consistent with the automatic prediction labels.”
How Image Degradations Affect Deep CNN-Based Face Recognition? (2016, methodology): “Finetuning has been done using a combination of the following datasets: CASIA-Webface [14], FaceScrub [15], and CACD [16].”
RotDCF: Decomposition of Convolutional Filters for Rotation-Equivariant Deep Networks (2018, methodology): “As a real-world example, we test RotDCF on the Facescrub dataset [24] containing over 100,000 face images of 530 people.”
Overcoming catastrophic forgetting with hard attention to the task (2018, methodology): “The considered data sets are: CIFAR10 and CIFAR100 (Krizhevsky, 2009), FaceScrub (Ng & Winkler, 2014), FashionMNIST (Xiao et al., 2017), NotMNIST (Bulatov, 2011), MNIST (Le-Cun et al., 1998), SVHN (Netzer et al., 2011), and Traf-ﬁcSigns (Stallkamp et al., 2011).”
Federated Continual Learning with Weighted Inter-client Transfer (2020, methodology): “2) NonIID-50 : We use the following eight benchmark datasets: MNIST (LeCun et al., 1998), CIFAR-10/-100 (Krizhevsky & Hinton, 2009), SVHN (Netzer et al., 2011), Fashion-MNIST (Xiao et al., 2017), Not-MNIST (Bulatov, 2011), FaceScrub (Ng & Winkler, 2014), and TrafﬁcSigns (Stal-lkamp et al., 2011).”
Uncertainty-guided Continual Learning with Bayesian Neural Networks (2019, methodology): “Furthermore, to have a better understanding of our method, we evaluate our approach on continually learning a sequence of 8 datasets with different distributions using the identical sequence as in (Serra et al., 2018), which includes FaceScrub (Ng & Winkler, 2014), MNIST, CIFAR100, NotMNIST (Bulatov”
Scalable and Order-robust Continual Learning with Additive Parameter Decomposition (2019, methodology): “Furthermore, we compare APD-Net against HAT (Serrà et al., 2018) on a sequence of 8 heterogeneous dataset including CIFAR-10, CIFAR-100, FaceScrub (Ng & Winkler, 2014), MNIST (LeCun et al., 1998), NotMNIST (Bulatov, 2011), FashionMNIST (Xiao et al., 2017), SVHN, and TrafﬁcSign (Stal-lkamp et al.,…”
SphereFace: Deep Hypersphere Embedding for Face Recognition (2017, methodology): “The probe set consists of two existing datasets: Facescrub [19] and FGNet.”
VGGFace2: A Dataset for Recognising Faces across Pose and Age (2017, methodology): “In order to study the effect of pose and age variations in recognising faces, the MegaFace challenge [12] uses the subsets of FaceScrub [14] containing 4, 000 images from 80 identities and FG-NET [15] containing 975 images from 82 identities for evaluation.”
MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition (2016, methodology): “Note that the query set in MegaFace are selected from images from FaceScrub [12] and FG-NET [13], which contains 530 and 82 persons respectively.”
CosFace: Large Margin Cosine Loss for Deep Face Recognition (2018, methodology): “In this study, we use the Facescrub dataset as the probe set to evaluate the performance of our approach on both Megaface Challenge 1 and Challenge 2.”
DA-GAN: Instance-Level Image Translation by Deep Attention Generative Adversarial Networks (2018, methodology): “We conduct these experiments on several datasets, including MNIST [18], CUB-200-2011 [38], SVHN [11], FaceScrub [23] and AnimePlanet 1.”