基于偶然不确定性的单人姿态估计模型测试时增强算法
CSTR:
作者:
作者单位:

(1. 中国科学院光束控制重点实验室, 成都 610209;2. 中国科学院光电技术研究所, 成都 610209;3. 中国科学院大学, 北京 100049)

作者简介:

李 杰(1995-),男,河北省邯郸市人,博士研究生,主要研究方向为人体姿态估计/关键点检测、目标检测与跟踪;

通讯作者:

中图分类号:

TP183

基金项目:


A Testing-Time-Augmentation Algorithm for Single Human Pose Estimation Based on Aleatoric Uncertainty
Author:
Affiliation:

(1. Key Lab. of Optical Beam Control Engineering, Chinese Academy of Sciences, Chengdu 610209, CHN;2. Institute of Optics and Electronics Technology, Chinese Academy of Sciences, Chengdu 610209, CHN;3. University of Chinese Academy of Sciences, Beijing 100049, CHN)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对现有单人姿态估计网络结果缺乏可靠性评估和鲁棒性保障等问题,提出了一种基于偶然不确定性的测试时增强方法。该方法首先利用随机并行的数据增强和模型推理得到多样化输出,随后通过计算该输出的偶然不确定性得到其可靠性评估,最后根据可靠性将该输出及其不确定性进行加权融合以得到更准确鲁棒的结果及其评估。在MPII数据集上的实验表明,该算法可即插即用地应用于任意现有单人姿态估计网络,从而得到更精确鲁棒的结果及其不确定性评估。

    Abstract:

    Aiming at the problem of lacking reliability evaluation and robustness guarantee in existing single-person pose estimation networks' results, a testing-time-augmentation (TTA) algorithm based on aleatoric uncertainty was proposed. In this TTA algorithm, diverse outputs were firstly obtained by stochastic parallel data augmentation and model inference. Then, the reliability evaluations of those outputs are acquired by calculating their aleatoric uncertainty. Finally, those outputs and their uncertainty were fused according to the reliabilities to obtain a more accurate and robust result as well as its evaluation. Experiments on the MPII dataset show that this algorithm can be applied to any existing single-person pose estimation network in a plug-and-play manner, leading to a more precise and robust result with its uncertainty evaluation.

    参考文献
    相似文献
    引证文献
引用本文

李杰,亓波,张建林.基于偶然不确定性的单人姿态估计模型测试时增强算法[J].半导体光电,2022,43(5):968-973. LI Jie, QI Bo, ZHANG Jianlin. A Testing-Time-Augmentation Algorithm for Single Human Pose Estimation Based on Aleatoric Uncertainty[J].,2022,43(5):968-973.

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2022-02-23
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2022-12-21
  • 出版日期:
文章二维码