Learning Face Representation from Scratch

By Dong Yi, Zhen Lei, Shengcai Liao, S. Li. 2014

2,202 citations recorded by the source corpus.

Pushing by big data and deep convolutional neural network (CNN), the performance of face recognition is becoming comparable to human. Using private large scale training datasets, several groups achieve very high performance on LFW, i.e., 97% to 99%. While there are many open source implementations of CNN, none of large scale face dataset is publicly available. The current situation in the field of face recognition is that data is more important than algorithm. To solve this problem, this paper proposes a semi-automatical way to collect face images from Internet and builds a large scale dataset containing about 10,000 subjects and 500,000 images, called CASIAWebFace. Based on the database, we use a 11-layer CNN to learn discriminative representation and obtain state-of-theart accuracy on LFW and YTF. The publication of CASIAWebFace will attract more research groups entering this field and accelerate the development of face recognition in the wild.

Subfields: Computer Vision and Pattern Recognition

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

FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age (2019, background): “To date, numerous large scale face image datasets [22, 31, 13, 69, 37, 24, 41, 68, 15, 27, 46, 8, 39] have been proposed and fostered research and development for automated face detection [35, 21], alignment [66, 44], recognition [57, 49], generation [67, 5, 26, 58], modiﬁcation [3, 32, 19], and…”
This Face Does Not Exist... But It Might Be Yours! Identity Leakage in Generative Models (2020, neutral): “3M) 512 Euclidean Distance ArcFace [7] ResNet-100 MS-Celeb-1M [12] (∼10M) 512 Cosine Distance SphereFace [19] ResNet-64 CASIA-Webface [30] (∼0.”
The Non-IID Data Quagmire of Decentralized Machine Learning (2019, methodology): “, 2016) over the CASIA-WebFace (Yi et al., 2014) dataset.”
Diversity in Faces (2019, background): “3(a) summarizes the feature distribution for age prediction in coding scheme 7, where we bin the age values into seven groups: [0, 1, 2, 3],[4, 5, 6, 7, 8, 9, 10, 11, 12],[13, 14, 15, 16, 17, 18, 19],[20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30],[31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44”
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].”
SphereFace: Deep Hypersphere Embedding for Face Recognition (2017, methodology): “CASIA-WebFace has 494,414 face images belonging to 10,575 different individuals.”
VGGFace2: A Dataset for Recognising Faces across Pose and Age (2017, methodology): “9 - - 2014 CASIA-WebFace [25] 10, 575 494, 414 2/46.”
The MegaFace Benchmark: 1 Million Faces for Recognition at Scale (2015, methodology): “JB algorithm was trained on the CASIA-WebFace dataset [36].”
Sampling Matters in Deep Embedding Learning (2017, methodology): “For verification, we train our model on the largest publicly available face dataset, CASIA-WebFace [40], and evaluate on the standard LFW [16] dataset.”
MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition (2016, methodology): “The CASIA-WebFace [16] is currently the largest dataset which is publicly available, with about 10K celebrities, and 500K images.”
CosFace: Large Margin Cosine Loss for Deep Face Recognition (2018, methodology): “Following the experimental setting in [23], we train a model with the guidance of the proposed LMCL on the CAISAWebFace[46] using the same 64-layer CNN architecture described in [23].”
Discrimination-aware Channel Pruning for Deep Neural Networks (2018, methodology): “We use CASIA-WebFace [49] (which consists of 494,414 face images from 10,575 individuals) for training.”
Web-scale training for face identification (2014, neutral): “We reiterate the importance of having a large scale training set in hand: training the initial net [22] using 500K images on CASIA [25], the largest public face dataset available today, obtains verification accuracy of only 93.”
Person Recognition in Personal Photo Collections (2015, methodology): “To measure the effect of training on larger data we consider fine-tuning using two open face recognition datasets: CASIA-WebFace (CASIA) [40] and the “Cross-Age Reference Coding Dataset” (CACD) [6].”
Quality-Agnostic Image Recognition via Invertible Decoder (2021, methodology): ““QualNet50-LM” is different from QualNet, which is described in Section 4.1. els trained for Table 4 where the models are trained with CASIA-WebFace [12].”
Range Loss for Deep Face Recognition with Long-Tailed Training Data (2016, methodology): “5M filtered data from MS-Celeb-1M [5] and CASIA-WebFace [34].”
Region Attention Networks for Pose and Occlusion Robust Facial Expression Recognition (2019, methodology): “Levi and Hassner [33] leverage the CASIA-WebFace[63] face recognition dataset to pretrain four different VGGNet[53] and GoogleNet[55].”
Angular Visual Hardness (2019, methodology): “For example, the CASIA-WebFace dataset [68] mostly contains white celebrities, so the neural network trained on CASIA-WebFace is highly biased against the other races.”
Additive Margin Softmax for Face Verification (2018, methodology): “Finally, we find 17 overlapped identities between CASIA-Webface and LFW, and 42 overlapped identities between CASIA-Webface and MegaFace set1.”