Learning Deep Face Representation

By Haoqiang Fan, Zhimin Cao, Yuning Jiang, Qi Yin, Chinchilla Doudou. 2014

113 citations recorded by the source corpus.

Face representation is a crucial step of face recognition systems. An optimal face representation should be discriminative, robust, compact, and very easy-to-implement. While numerous hand-crafted and learning-based representations have been proposed, considerable room for improvement is still present. In this paper, we present a very easy-to-implement deep learning framework for face representation. Our method bases on a new structure of deep network (called Pyramid CNN). The proposed Pyramid CNN adopts a greedy-filter-and-down-sample operation, which enables the training procedure to be very fast and computation-efficient. In addition, the structure of Pyramid CNN can naturally incorporate feature sharing across multi-scale face representations, increasing the discriminative ability of resulting representation. Our basic network is capable of achieving high recognition accuracy ($85.8\%$ on LFW benchmark) with only 8 dimension representation. When extended to feature-sharing Pyramid CNN, our system achieves the state-of-the-art performance ($97.3\%$) on LFW benchmark. We also introduce a new benchmark of realistic face images on social network and validate our proposed representation has a good ability of generalization.

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

Learning feed-forward one-shot learners (2016, background): “Training is usually performed by classifying pairs according to distance [5], or by enforcing a distance ranking with a triplet loss [16].”
Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning (2017, background): “Recently, various deep neural network architectures are proposed for face recognition tasks, and achieve impressive performance [58], [63], [55], [62], [26], [74].”
Accessorize to a Crime: Real and Stealthy Attacks on State-of-the-Art Face Recognition (2016, methodology): “It has been shown to achieve accuracy over 97.3% on LFW [8].”
Surpassing Human-Level Face Verification Performance on LFW with GaussianFace (2014, background): “Using it also allows us to compare directly with other existing face veriﬁcation methods [9, 5, 14, 47, 13, 59, 1, 20, 16].”
Who Makes Trends? Understanding Demographic Biases in Crowdsourced Recommendations (2017, background): “More broadly, our work offers a new perspective to the growing debate about fairness, bias, and transparency of decisions taken by algorithms operating over big crowd-sourced data (Zafar et al. 2015; Chakraborty et al. 2016; Kroll 2015).”
Twitter's Glass Ceiling: The Effect of Perceived Gender on Online Visibility (2016, methodology): “We labeled users as ambiguous if the confidence level reported by Face++ was below 95%.”
Recover Canonical-View Faces in the Wild with Deep Neural Networks (2014, methodology): “…[9] 96.33 FR+FCNt (whol巳 fac巳+components) 96 .4 5 PLDA [25] 90.07 Joint Bayesian [12] 90.90 Fisher Vector Faces [5] 93.03 High-dimLBP[11] 93.18 FR+FCN (whole fac巳) 93.65 FR+FCN (whol巳 fac巳+components) 94.38 Face++ [16] 97.27 3) the size of the dataset is not 巳nough for a deep learning-based method.”
Demographics of News Sharing in the U.S. Twittersphere (2017, methodology): “…work, we use the profile picture’s URLs of all users in our dataset and use the Face++ API , a face recognition platform based on deep learning [43], to infer the gender (i.e., male or female), race (limited to Asian, Black , and White) and age information from the recognized faces in the…”
Labeled Faces in the Wild: A Survey (2016, methodology): “More recently, Taigman et al. [97] developed a frontalization method that contributed a modest improvement to accuracy on LFW, although most of their gains are attributable to their CNN architecture and the large training sets.”
Speaker diarization through speaker embeddings (2015, background): “Representation learning has led to interesting improvements in various domains, such as face recognition [9], text modeling [10] or speech recognition [11].”
Face Recognition and Retrieval Using Cross-Age Reference Coding With Cross-Age Celebrity Dataset (2015, methodology): “To further understand the dataset, we also run race and gender detection using a commercial system [37] to analyze the gender and race distributions of the datasets.”
PCANet: A Simple Deep Learning Baseline for Image Classification? (2014, methodology): “In preparation of this paper, we are aware of two concurrent works [34], [35] that employ ConvNet for LFW face verification.”
Energy-Fluctuated Multiscale Feature Learning With Deep ConvNet for Intelligent Spindle Bearing Fault Diagnosis (2017, background): “More recently, deep learning (DL), as a new research hotspot of machine learning, has been achieved state-of-art in many advanced fields, including face verification and speech recognition, and many powerful features have been proposed, such as deep face [11], deepID [12], and DeepSpeech [13].”