Shiyao Huang

Shiyao Huang

Auto-generated profile, based on publicly available information

Shiyao Huang

SenseTime Research

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
5 October 2018
Verified
Book Chapter
Closed access
PDF file
FWFei Wang
Fei Wang
LCLiren Chen
Liren Chen
CLCheng Li
Cheng Li
SHShiyao Huang
Shiyao Huang
YCYanjie Chen
Yanjie Chen
CLChen Change Loy
Chen Change Loy

1 December 2016
Verified
Conference Paper
Closed access
SYSen Yang
Sen Yang
JRJiangpeng Rong
Jiangpeng Rong
SHShiyao Huang
Shiyao Huang
ZSZeyu Shang
Zeyu Shang
YSYongjie Shi
Yongjie Shi
HZHongbin Zha
Hongbin Zha
Radial lens distortion often exists in common cameras, which makes the taken images violate the assumption of pinhole camera model. The prestigious division model for radial distortion with just a single parameter is presented by Fitzgibbon. Lines in 3D space may be projected onto curves in the image plane due to radial distortion. Alemán et al. utilized Hough transform with three parameters, one parameter coming from the division model, and the other two arising from the corrected image line, to automatically and simultaneously correct radial lens distortion and detect multiple image lines. The score function used in the method of Alemán et al. is the total weighted length of all peaks greater than some threshold in Hough space, namely, the image lines greater than some preset length. However, the corrected results from some distorted images are not very satisfactory using the method proposed by Alemán et al. In this paper, we present a novel score function which combines statistical quantities with grouping based on vanishing points from multiple line images. The experimental results validated the proposed method.

31 July 2018 in arXiv
Verified
Preprint
Open access
PDF file
FWFei Wang
Fei Wang
LCLiren Chen
Liren Chen
CLCheng Li
Cheng Li
SHShiyao Huang
Shiyao Huang
YCYanjie Chen
Yanjie Chen
CLChen Change Loy
Chen Change Loy
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.