MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition
ECCV, 2016.
EI
Weibo:
Abstract:
In this paper, we design a benchmark task and provide the associated datasets for recognizing face images and link them to corresponding entity keys in a knowledge base. More specifically, we propose a benchmark task to recognize one million celebrities from their face images, by using all the possibly collected face images of this indivi...More
Code:
Data:
Introduction
- The authors design a benchmark task as to recognize one million celebrities from their face images and identify them by linking to the unique entity keys in a knowledge base.
- The current face identification task mainly focuses on finding similar images for the input image, rather than answering questions such as “who is in the image?” and “if it is Anne in the image, which Anne?”.
- This lacks an important step of “recognizing”.
Highlights
- In this paper, we design a benchmark task as to recognize one million celebrities from their face images and identify them by linking to the unique entity keys in a knowledge base
- – We provide the following datasets,2 – One million celebrities selected from freebase with corresponding entity keys, and a snapshot for freebase data dumps; – Manually labeled measurement set with carefully designed evaluation protocol; – A large scale training dataset, with face region cropped and aligned
- We provide concrete measurement set for people to evaluate the model performance and provide, to the best of our knowledge, the largest training dataset to facilitate research in the area
- The images in our training dataset are associated with entity keys in knowledge base, of which the gender information could be retrieved
- People could train a robust gender classifier for the face images in the wild based on this large scale training data
- We look forward to exciting research inspired by our training dataset and benchmark task
Results
- Evaluation Protocol
The authors evaluate the performance of the proposed recognition task in terms of precision and coverage using the settings described as follows.
Setup The authors setup the evaluation protocol as follows. - The chance to include the measurement images in the training set is relatively low, as long as the celebrity list in the measurement set is hidden
- This is different from most of the existing face recognition benchmark tasks, in which the measurement set is published and targeted on a small group of people.
Conclusion
- Discussion and Future work
In this paper, the authors have defined a benchmark task which is to recognize one million celebrities in the world from their face images, and link the face to a corresponding entity key in a knowledge base. - People could adopt one of the cutting-edge unsupervised/semisupervised clustering algorithms [21] [22] [23] [24] on the training dataset, and/or develop new algorithms which can accurately locate and remove outliers in a large, real dataset
- Another interesting topic is the to build estimators to predict a person’s properties from his/her face images.
- The authors look forward to exciting research inspired by the training dataset and benchmark task
Summary
Introduction:
The authors design a benchmark task as to recognize one million celebrities from their face images and identify them by linking to the unique entity keys in a knowledge base.- The current face identification task mainly focuses on finding similar images for the input image, rather than answering questions such as “who is in the image?” and “if it is Anne in the image, which Anne?”.
- This lacks an important step of “recognizing”.
Results:
Evaluation Protocol
The authors evaluate the performance of the proposed recognition task in terms of precision and coverage using the settings described as follows.
Setup The authors setup the evaluation protocol as follows.- The chance to include the measurement images in the training set is relatively low, as long as the celebrity list in the measurement set is hidden
- This is different from most of the existing face recognition benchmark tasks, in which the measurement set is published and targeted on a small group of people.
Conclusion:
Discussion and Future work
In this paper, the authors have defined a benchmark task which is to recognize one million celebrities in the world from their face images, and link the face to a corresponding entity key in a knowledge base.- People could adopt one of the cutting-edge unsupervised/semisupervised clustering algorithms [21] [22] [23] [24] on the training dataset, and/or develop new algorithms which can accurately locate and remove outliers in a large, real dataset
- Another interesting topic is the to build estimators to predict a person’s properties from his/her face images.
- The authors look forward to exciting research inspired by the training dataset and benchmark task
Tables
- Table1: Face recognition datasets
- Table2: Experimental results on the 500 published celebrities
Related work
- Typically, there are two types of tasks for face recognition. One is very wellstudied, called face verification, which is to determine whether two given face images belong to the same person. Face verification has been heavily investigated. One of the most widely used measurement sets for verification is Labeled Faces in the Wild (LFW) in [7,8], which provides 3000 matched face image pairs and 3000 mismatched face image pairs, and allows researchers to report verification accuracy with different settings. The best performance on LFW datasets has been frequently updated in the past several years. Especially, with the “unrestricted, labeled outside data” setting, multiple research groups have claimed higher accuracy than human performance for verification task on LFW [4,9].
Reference
- Guo, Y., Zhang, L., Hu, Y., He, X., Gao, J.: MS-Celeb-1M: Challenge of recognizing one million celebrities in the real world. In: IS&T International Symposium on Electronic Imaging. (2016)
- Taigman, Y., Yang, M., Ranzato, M., Wolf, L.: Deepface: Closing the gap to human-level performance in face verification. In: Proc. of IEEE Computer Soc. Conf. on Computer Vision and Pattern Recognition (CVPR). (June 2014)
- Taigman, Y., Yang, M., Ranzato, M., Wolf, L.: Web-scale training for face identification. In: Proc. of IEEE Computer Soc. Conf. on Computer Vision and Pattern Recognition (CVPR), IEEE (2015) 2746–2754
- Schroff, F., Kalenichenko, D., Philbin, J.: Facenet: A unified embedding for face recognition and clustering. In: Proc. of IEEE Computer Soc. Conf. on Computer Vision and Pattern Recognition (CVPR). (June 2015)
- Google: Freebase data dumps. https://developers.google.com/freebase/data (2015)
- Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., Berg, A.C., Fei-Fei, L.: ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision (IJCV) 115(3) (2015) 211–252
- Huang, G.B., Ramesh, M., Berg, T., Learned-Miller, E.: Labeled faces in the wild: A database for studying face recognition in unconstrained environments. Technical Report 07-49, University of Massachusetts, Amherst (October 2007)
- Huang, G.B., Learned-Miller, E.: Labeled faces in the wild: Updates and new reporting procedures. Technical Report UM-CS-2014-003, University of Massachusetts, Amherst (May 2014)
- Sun, Y., Wang, X., Tang, X.: DeepID3: Face recognition with very deep neural networks. arXiv preprint arXiv:1502.00873 (2014)
- Fan, H., Yang, M., Cao, Z., Jiang, Y., Yin, Q.: Learning compact face representation: Packing a face into an int32. In: Proc. of ACM Int’l Conf. on Multimedia, ACM (2014) 933–936
- Kemelmacher-Shlizerman, I., Seitz, S., Miller, D., Brossard, E.: The MegaFace benchmark: 1 million faces for recognition at scale. ArXiv e-prints (2015)
- Ng, H.W., Winkler, S.: A data-driven approach to cleaning large face datasets. In: Proc. of IEEE Int’l Conf. on Image Proc. (ICIP). (Oct 2014)
- Panis, G., Lanitis, A.: An overview of research activities in facial age estimation using the FG-NET aging database. In: Proc. of the European Conf. on Computer Vision (ECCV) Workshops. (2014)
- Wolf, L., Hassner, T., Maoz, I.: Face recognition in unconstrained videos with matched background similarity. In: Proc. of IEEE Computer Soc. Conf. on Computer Vision and Pattern Recognition (CVPR). (2011)
- Sun, Y., Wang, X., Tang, X.: Deep learning face representation from predicting 10,000 classes. In: Proc. of IEEE Computer Soc. Conf. on Computer Vision and Pattern Recognition (CVPR). (June 2014)
- Yi, D., Lei, Z., Liao, S., Li, S.Z.: Learning face representation from scratch. arXiv preprint arXiv:1411.7923 (2014)
- Klare, B.F., Klein, B., Taborsky, E., Blanton, A., Cheney, J., Allen, K., Grother, P., Mah, A., Jain, A.K.: Pushing the frontiers of unconstrained face detection and recognition: Iarpa janus benchmark a. In: Proc. of IEEE Computer Soc. Conf. on Computer Vision and Pattern Recognition (CVPR). (June 2015)
- Parkhi, O.M., Vedaldi, A., Zisserman, A.: Deep face recognition. In: Proceedings of the British Machine Vision Conference (BMVC). (2015)
- Eastman, G.: Camera. US Patent 388850 A (1888)
- Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. In: Advances in Neural Information Processing Systems (NIPS), MIT Press (2012) 1097–1105 21. Ng, A.Y., Jordan, M.I., Weiss, Y.: On spectral clustering: Analysis and an algorithm. In: Advances in Neural Information Processing Systems (NIPS), MIT Press (2001) 849–856 22.
- Belkin, M., Niyogi, P.: Semi-supervised learning on riemannian manifolds. Journal of Machine Learning 56(1-3) (June 2004) 209–239 23.
- Zhu, X., Ghahramani, Z., Lafferty, J.: Semi-supervised learning using gaussian fields and harmonic functions. In: Proc. of Int’l Conf. on Machine Learning. (2003) 912–919 24.
- Zhou, D., Bousquet, O., Lal, T.N., Weston, J., Schlkopf, B.: Learning with local and global consistency. In: Advances in Neural Information Processing Systems (NIPS), MIT Press (2004) 321–328
Full Text
Tags
Comments