Do We Really Need to Collect Millions of Faces for Effective Face Recognition?

By Iacopo Masi, Anh Tuan Tran, Tal Hassner, Jatuporn Toy Leksut, Gérard G. Medioni. European Conference on Computer Vision, 2016

379 citations recorded by the source corpus.

Face recognition capabilities have recently made extraordinary leaps. Though this progress is at least partially due to ballooning training set sizes – huge numbers of face images downloaded and labeled for identity – it is not clear if the formidable task of collecting so many images is truly necessary. We propose a far more accessible means of increasing training data sizes for face recognition systems: Domain specific data augmentation. We describe novel methods of enriching an existing dataset with important facial appearance variations by manipulating the faces it contains. This synthesis is also used when matching query images represented by standard convolutional neural networks. The effect of training and testing with synthesized images is tested on the LFW and IJB-A (verification and identification) benchmarks and Janus CS2. The performances obtained by our approach match state of the art results reported by systems trained on millions of downloaded images.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/376973681a3e717736625bd4e680515b909cf339

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

Citations in the corpus

Improving Deep Learning using Generic Data Augmentation (2017, background): “For example, Masi et al. [Masi et al. , 2016] developed a facial recognition system us-ing synthesized faces with different poses and facial expressions to increase appearance variability enabling comparable task performance to state of the art facial recognition systems using less training data.”
Neural Face Editing with Intrinsic Image Disentangling (2017, methodology): “[23] used face fitting using morphable models similar to our approach, but used the resulting 3D faces to generate more data for traditional recognition network training.”
Learning Pose-Aware Models for Pose-Invariant Face Recognition in the Wild (2019, methodology): “Similar to [15], [16], PAMs also allow us to overcome a main limitation of the pose-deﬁcient training set, resulting in a better trained system.”
Past, Present, and Future of Face Recognition: A Review (2020, methodology): “[123] (2016) have augmented their training data by generating new facial images with specific appearance variations, including shape, pose, and expression using a 3D generic face.”
Wing Loss for Robust Facial Landmark Localisation with Convolutional Neural Networks (2017, methodology): “In addition, 3D face models have also been widely used to synthesise additional 2D face images with pose variations for the training of a pose-invariant system [41, 17, 76].”
Data Augmentation in Classification and Segmentation: A Survey and New Strategies (2023, methodology): “For example, Mash et al. [39] used a ﬁne-grained dataset of ten classes to test a variety of geometric augmentation methods for the task of aircraft classiﬁcation.”
Frankenstein: Learning Deep Face Representations Using Small Data (2016, neutral): “Similar data synthesis techniques are also used for poserobust face recognition [30].”
Extreme 3D Face Reconstruction: Looking Past Occlusions (2017, methodology): “For this purpose, we use the publicly available deep models trained for face recognition by [28].”
On Face Segmentation, Face Swapping, and Face Perception (2017, methodology): “We use the ResFace101 [33] face recognition system to test if faces remain recognizable after swapping.”
A survey on face data augmentation for the training of deep neural networks (2019, methodology): “Meanwhile many works used 3D face models for face pose translation [6, 9, 34–37].”
Improving Deep Learning with Generic Data Augmentation (2018, background): “For example, Masi et al. [13] developed a facial recognition system using synthesized faces with different poses and facial expressions to increase appearance variability enabling comparable task performance to state of the art facial recognition systems using less training data.”
Marginal Loss for Deep Face Recognition (2017, methodology): “In Table 1, we compare the proposed marginal loss method against many existing state-of-the-art models, in- cluding DeepID [20], VGG Face [17], Deep Face [24], Fusion [25], FaceNet [18], Baidu [13], Center Loss [27], Range Loss [31], Multibatch [23], Aug [14], and our baseline softmax loss.”
Unequal-Training for Deep Face Recognition With Long-Tailed Noisy Data (2019, neutral): “Augumentation [23] 1 19 1.”