Deep Learning Face Representation by Joint Identification-Verification

By Yi Sun, Yuheng Chen, Xiaogang Wang, Xiaoou Tang. 2014

2,318 citations recorded by the source corpus.

The key challenge of face recognition is to develop effective feature representations for reducing intra-personal variations while enlarging inter-personal differences. In this paper, we show that it can be well solved with deep learning and using both face identification and verification signals as supervision. The Deep IDentification-verification features (DeepID2) are learned with carefully designed deep convolutional networks. The face identification task increases the inter-personal variations by drawing DeepID2 features extracted from different identities apart, while the face verification task reduces the intra-personal variations by pulling DeepID2 features extracted from the same identity together, both of which are essential to face recognition. The learned DeepID2 features can be well generalized to new identities unseen in the training data. On the challenging LFW dataset [11], 99.15% face verification accuracy is achieved. Compared with the best previous deep learning result [20] on LFW, the error rate has been significantly reduced by 67%.

Subfields: Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning, 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 Representation & Self-Supervised Learning

In Signal Processing

Citations in the corpus

Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification (2015, background): “Convolutional neural networks (CNNs) [19, 18] have demonstrated recognition accuracy better than or comparable to humans in several visual recognition tasks, including recognizing traffic signs [3], faces [34, 32], and handwritten digits [3, 36].”
FaceNet: A unified embedding for face recognition and clustering (2015, methodology): “the one using pairs of positives and negatives, as used in [14] Eq.”
Deep Learning Face Attributes in the Wild (2014, methodology): “Large-scale Data Collection We construct two face attribute datasets, namely CelebA and LFWA, by labeling images selected from two challenging face datasets, CelebFaces [26] and LFW [12].”
Semantic Image Segmentation via Deep Parsing Network (2015, background): “• b12 As listed in Table 1 (b), ‘lconv’ in b12 indicates a locally convolutional layer, which is widely used in face recognition [33, 35] to capture different information from different facial positions.”
Exploring the Encoding Layer and Loss Function in End-to-End Speaker and Language Recognition System (2018, background): “…function is needed to have better generalization than closed-set identiﬁcation: In [21, 22], similar pairwise loss such as contrastive loss [30, 31] or triplet loss [32] is adopted for open-set speaker ver-Front-end However, a neural network trained with pairwise loss requires carefully…”
SphereFace: Deep Hypersphere Embedding for Face Recognition (2017, methodology): “To address this, some methods combine softmax loss with contrastive loss [25, 28] or center loss [34] to enhance the discrimination power of features.”
Image Super-Resolution Using Deep Convolutional Networks (2014, background): “First, its structure is intentionally designed with simplicity in mind, and yet provides superior accuracy2 compared with state-of-the-art example-based methods.”
Large-Margin Softmax Loss for Convolutional Neural Networks (2016, background): “To enhance the intra-class compactness and inter-class separability, (Sun et al., 2014) trains the CNN with the combination of softmax loss and contrastive loss.”
Sampling Matters in Deep Embedding Learning (2017, background): “DeepID2 [30] and DeepID3 [31] use 25 networks on 25 face regions based on positions of facial landmarks.”
Learning Face Representation from Scratch (2014, methodology): “Our network integrates the most popular components in the recent works, such as ReLU neuron [20], dropout [6], low dimensional representation, identification + verification cost function [24], small filter and very deep architecture [23].”
CosFace: Large Margin Cosine Loss for Deep Face Recognition (2018, methodology): “DeepID2 [30] employs identification and verification signals to achieve better feature embedding.”
Naive-Deep Face Recognition: Touching the Limit of LFW Benchmark or Not? (2015, background): “Extensive works have been done to push the accuracy limit on it [3, 16, 4, 1, 2, 5, 11, 10, 12, 14, 13, 17, 9].”
End-to-End Learning of Deep Visual Representations for Image Retrieval (2016, background): “Siamese networks have performed well for metric learning (Song et al, 2016), dimensionality reduction (Hadsell et al, 2006), learning image descriptors (SimoSerra et al, 2015), and performing face identification (Chopra et al, 2005; Hu et al, 2014; Sun et al, 2014).”
Web-scale training for face identification (2014, background): “In fact, recent contributions [20, 22, 19] reported nearly human level performance on the LFW verification task using deep neural networks, but no test is reported on the probe-gallery identification task.”
A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects (2020, methodology): “It was first used in SiameseNet [67] and later was deployed in DeepID2 [68], DeepID2+ [69], and DeepID3 [70].”
Deep Metric Learning with Hierarchical Triplet Loss (2018, background): “…key-point descriptor learning [12,24], face veriﬁcation [22,20], person re-identiﬁcation [30,23], object tracking [29] and etc. Recently, there is a number of widely-used loss functions developed for deep metric learning, such as contrastive loss [27,6], triplet loss [22] and quadruplet loss [5].”
Improving Person Re-identification by Attribute and Identity Learning (2017, methodology): “We first describe two baselines in Section 4.1 and then the APR network in Section 4.2”