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. = equal contribution arXiv:1807.11649v1 [cs.CV] 31 Jul 2018 2 Fei Wang et al .

A novel zirconia-based HPLC packing material, ZrO2/SiO2, which consists of micrometer-sized silica spheres as core and nanometer-sized zirconia particles as surface coating, was prepared by a layer-by-layer self-assembly technique. The material exhibits favorable characteristics for HPLC applications, including high surface area and pore volume, good pore structure, narrow particle size, and pore size distribution. Not only the support ZrO2/SiO2 but also the stationary-phase C18 bonded ZrO2/SiO2 exhibits excellent chemical stability. In addition, good permeability was observed for both of them. High specific area surface and good permeability of ZrO2/SiO2 permit a high loading amount of chiral polymer on it and greatly improved the enantioselectivity and resolution for some chiral separations.

No abstract