The Do’s and Don’ts for CNN-Based Face Verification
Authors:
- Ankan BansalABAnkan Bansal
- Carlos CastilloCCCarlos Castillo
- Rajeev RanjanRRRajeev Ranjan
- Rama ChellappaRCRama Chellappa
- 1 October 2017
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2545-2554
Abstract
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) Is alignment needed for face recognition? We address these questions by training CNNs using CASIA-WebFace, UMD-Faces, and a new video dataset and testing on YouTube-Faces, IJB-A and a disjoint portion of UMD-Faces datasets. Our new data set, which will be made publicly available, has 22,075 videos and 3,735,476 human annotated frames extracted from them.Keywords
This publication has 20 references indexed in Scilit:
- Triplet probabilistic embedding for face verification and clustering2016
- The MegaFace Benchmark: 1 Million Faces for Recognition at Scale2016
- FaceNet: A unified embedding for face recognition and clustering2015
- Pushing the frontiers of unconstrained face detection and recognition: IARPA Janus Benchmark A2015
- Caffe2014
- One millisecond face alignment with an ensemble of regression trees2014
- Hybrid Deep Learning for Face Verification2013
- Face recognition in unconstrained videos with matched background similarity2011
- Attribute and simile classifiers for face verification2009
- Robust Real-Time Face DetectionInternational Journal of Computer Vision, 2004