Low-resolution human pose estimation
作者：
Highlights：
• Analyzing effect of low-resolution input on human pose estimation.
• Exploring the significant advantages of offset-based methods in the low-resolution setting.
• Gaussian offset weighing guides the learning of offset fields.
• Gaussian heatmap weighing represents accurate human joint positions.
• Two-stage training strategy is proposed for robust guiding information.
摘要
•Analyzing effect of low-resolution input on human pose estimation.•Exploring the significant advantages of offset-based methods in the low-resolution setting.•Gaussian offset weighing guides the learning of offset fields.•Gaussian heatmap weighing represents accurate human joint positions.•Two-stage training strategy is proposed for robust guiding information.
论文关键词：Human pose estimation,Low resolution image,Heatmap learning,Offset learning,Quantization error
论文评审过程：Received 12 January 2021, Revised 6 February 2022, Accepted 7 February 2022, Available online 9 February 2022, Version of Record 20 February 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108579
