The Do’s and Don’ts for CNN-Based Face Verification

@article{Bansal2017TheDA,
  title={The Do’s and Don’ts for CNN-Based Face Verification},
  author={Ankan Bansal and Carlos D. Castillo and Rajeev Ranjan and Rama Chellappa},
  journal={2017 IEEE International Conference on Computer Vision Workshops (ICCVW)},
  year={2017},
  pages={2545-2554}
}
While the research community appears to have developed a consensus on the methods of acquiring annotated data, design and training of CNNs, many questions still remain to be answered. In this paper, we explore the following questions that are critical to face recognition research: (i) Can we train on still images and expect the systems to work on videos? (ii) Are deeper datasets better than wider datasets? (iii) Does adding label noise lead to improvement in performance of deep networks? (iv… 
View on IEEE
arxiv.org
83 Citations
Highly Influential Citations
7
Background Citations
37
Methods Citations
22
Results Citations
1

83 Citations

To Frontalize or Not to Frontalize: Do We Really Need Elaborate Pre-processing to Improve Face Recognition?
TLDR
A number of facial landmarking algorithms and a popular frontalization method are evaluated and a new, automatic, single-image frontalization scheme is introduced that exceeds the performance of the reference frontalization algorithm for video-to-video face matching on the Point and Shoot Challenge dataset.
A Light CNN for Deep Face Representation With Noisy Labels
TLDR
Experimental results show that the proposed framework can utilize large-scale noisy data to learn a Light model that is efficient in computational costs and storage spaces and achieves state-of-the-art results on various face benchmarks without fine-tuning.
The Do’s and Don’ts for Increasing the Accuracy of Face Recognition on VGGFace2 Dataset
TLDR
In LFW, one of the widely used datasets in the literature, an accuracy rate of 100% has been achieved by exceeding the highest accuracy achieved so far and all images in the LFW database have been recognized without any problems.
Exploring the Effects of Frontalization and Data Synthesis on Face Recognition
TLDR
A novel face synthesis method for augmentation of existing face image datasets that reduces overfitting, which in turn, can enhance the face representation capability of a CNN.
A Fast and Accurate System for Face Detection, Identification, and Verification
TLDR
A novel face detector, deep pyramid single shot face detector (DPSSD), which is fast and detects faces with large scale variations (especially tiny faces), and a new loss function, called crystal loss, for the tasks of face verification and identification.
Crystal Loss and Quality Pooling for Unconstrained Face Verification and Recognition
TLDR
This paper proposes a new loss function, called Crystal Loss, that restricts the features to lie on a hypersphere of a fixed radius and shows that integrating this simple step in the training pipeline significantly improves the performance of face verification and recognition systems.
VGGFace2: A Dataset for Recognising Faces across Pose and Age
TLDR
A new large-scale face dataset named VGGFace2 is introduced, which contains 3.31 million images of 9131 subjects, with an average of 362.6 images for each subject, and the automated and manual filtering stages to ensure a high accuracy for the images of each identity are described.
Deep Learning for Understanding Faces
TLDR
This paper discusses different modules involved in designing an automatic face recognition system and the role of deep learning for each of them and some open issues regarding DCNNs for face recognition problems are discussed.
Investigating Nuisances in DCNN-Based Face Recognition
TLDR
A thorough analysis of several aspects that impact on the use of DCNN for face recognition shows how competitive performance with respect to the state of the art can be reached even with standard DCNN architectures and pipeline.
A Comparative Study of Face Re-identification Systems under Real-World Conditions
TLDR
It is shown that State-of-the-Art systems are still not fully robust, and give a much poorer performance when used on real-world still-to-video data.
...
...

References

SHOWING 1-10 OF 43 REFERENCES
UMDFaces: An annotated face dataset for training deep networks
TLDR
A new face dataset, called UMDFaces, is introduced, which has 367,888 annotated faces of 8,277 subjects and a new face recognition evaluation protocol is introduced which will help advance the state-of-the-art in this area.
Learning Face Representation from Scratch
TLDR
A semi-automatical way to collect face images from Internet is proposed and a large scale dataset containing about 10,000 subjects and 500,000 images, called CASIAWebFace is built, based on which a 11-layer CNN is used to learn discriminative representation and obtain state-of-theart accuracy on LFW and YTF.
Deep Face Recognition
TLDR
It is shown how a very large scale dataset can be assembled by a combination of automation and human in the loop, and the trade off between data purity and time is discussed.
Triplet probabilistic embedding for face verification and clustering
TLDR
This paper proposes an approach that couples a deep CNN-based approach with a low-dimensional discriminative embedding step, learned using triplet probability constraints to address the unconstrained face verification problem.
Unconstrained face verification using deep CNN features
TLDR
An algorithm for unconstrained face verification based on deep convolutional features and evaluate it on the newly released IARPA Janus Benchmark A (IJB-A) dataset as well as on the traditional Labeled Face in the Wild (LFW) dataset.
Deep Learning Face Attributes in the Wild
TLDR
A novel deep learning framework for attribute prediction in the wild that cascades two CNNs, LNet and ANet, which are fine-tuned jointly with attribute tags, but pre-trained differently.
Face recognition in unconstrained videos with matched background similarity
TLDR
A comprehensive database of labeled videos of faces in challenging, uncontrolled conditions, the ‘YouTube Faces’ database, along with benchmark, pair-matching tests are presented and a novel set-to-set similarity measure, the Matched Background Similarity (MBGS), is described.
MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition
TLDR
A benchmark task to recognize one million celebrities from their face images, by using all the possibly collected face images of this individual on the web as training data, which could lead to one of the largest classification problems in computer vision.
WIDER FACE: A Face Detection Benchmark
TLDR
There is a gap between current face detection performance and the real world requirements, and the WIDER FACE dataset, which is 10 times larger than existing datasets is introduced, which contains rich annotations, including occlusions, poses, event categories, and face bounding boxes.
DeepFace: Closing the Gap to Human-Level Performance in Face Verification
TLDR
This work revisits both the alignment step and the representation step by employing explicit 3D face modeling in order to apply a piecewise affine transformation, and derive a face representation from a nine-layer deep neural network.
...
...