The Devil of Face Recognition is in the Noise
arXiv:1807.11649
Abstract
The growing scale of face recognition datasets empowers us to train strong convolutional networks for face recognition. While a variety of architectures and loss functions have been devised, we still have a limited understanding of the source and consequence of label noise inherent in existing datasets. We make the following contributions: 1) We contribute cleaned subsets of popular face databases, i.e., MegaFace and MS-Celeb-1M datasets, and build a new large-scale noise-controlled IMDb-Face dataset. 2) With the original datasets and cleaned subsets, we profile and analyze label noise properties of MegaFace and MS-Celeb-1M. We show that a few orders more samples are needed to achieve the same accuracy yielded by a clean subset. 3) We study the association between different types of noise, i.e., label flips and outliers, with the accuracy of face recognition models. 4) We investigate ways to improve data cleanliness, including a comprehensive user study on the influence of data labeling strategies to annotation accuracy. The IMDb-Face dataset has been released on https://github.com/fwang91/IMDb-Face.
arXiv commentaccepted to ECCV'18
References in corpus (4)
Cited by in corpus (9)
- Knowledge Distillation via Route Constrained Optimization
- Masked Face Recognition Challenge: The WebFace260M Track Report
- Inducing Predictive Uncertainty Estimation for Face Recognition
- Mis-classified Vector Guided Softmax Loss for Face Recognition
- Towards Robustness to Label Noise in Text Classification via Noise Modeling
- The Resistance to Label Noise in K-NN and DNN Depends on its Concentration
- Correlation Congruence for Knowledge Distillation
- Learning Hierarchical Graph Neural Networks for Image Clustering
- More Information Supervised Probabilistic Deep Face Embedding Learning