106点面部关键点定位的重大挑战
计算机视觉与模式识别
2019-07-25 v3
摘要
面部关键点定位在众多人脸相关应用中是非常关键的步骤，如人脸识别、面部姿态估计、人脸图像合成等。然而，以往的面部关键点定位竞赛（即 300-W、300-VW 和 Menpo 挑战赛）旨在预测 68 点关键点，不足以刻画面部部件的结构。为克服此问题，我们构建了一个具有挑战性的数据集，名为 JD-landmark。每幅图像均手动标注了 106 点关键点。该数据集涵盖姿态与表情的大幅变化，给准确预测关键点带来诸多困难。我们结合 IEEE 国际多媒体与博览会 (ICME) 2019 在该数据集上举办了 106 点面部关键点定位竞赛1。该竞赛旨在发现有效且鲁棒的面部关键点定位方法。
引用
@article{arxiv.1905.03469,
title = {Grand Challenge of 106-Point Facial Landmark Localization},
author = {Yinglu Liu and Hao Shen and Yue Si and Xiaobo Wang and Xiangyu Zhu and Hailin Shi and Zhibin Hong and Hanqi Guo and Ziyuan Guo and Yanqin Chen and Bi Li and Teng Xi and Jun Yu and Haonian Xie and Guochen Xie and Mengyan Li and Qing Lu and Zengfu Wang and Shenqi Lai and Zhenhua Chai and Xiaoming Wei},
journal= {arXiv preprint arXiv:1905.03469},
year = {2019}
}
备注
This paper is accepted at ICME2019 Grand Challenge. The JD-landmark dataset has been released and can be downloaded from https://sites.google.com/view/hailin-shi