返回
Channel sifted model for pose estimation
DOI:10.1007/s10489-022-04091-1.png)
摘要
En 中文
In recent years, human pose estimation, also called keypoint estimation, has been a research spot in the field of computer vision. However, most methods focus on how to achieve high accuracy and ignore the computational cost. These methods usually use deeper layers and more complex structures to construct their networks result in a surge in computational cost, so it is difficult to apply them in real environments with real-time requirements. Some methods try to reduce the computational cost by decreasing precision of parameter or using simple structures. However, they can only achieve low performance, so it is difficult to apply in the real environments. Inspired by lightweight network design, this paper proposes a channel sifted method for human pose estimation, which is called the channel sifted network (CSN). A lightweight ResNet is used as the backbone, and a channel scoring module (CSM) is applied. Both parts aim to balance the computational cost and prediction accuracy of the network. The lower the computational cost, the faster the inference speed. In the experimental part, we first train and test the network on the COCO2017 dataset and MPII dataset respectively, and then demonstrate the effectiveness of lightweight backbone and CSM in the network through ablation study. Compared with other lightweight models, CSN achieves higher performance.
Keyword:
Pose estimation
Channel scoring module
Keypoint estimation
Channel sifted network
期刊
IF:
3.5
论文数:
7.6K
被引数:
1.7W
机构
暂无机构信息
引用论文
Thermal comfort, perceived air quality, and cognitive performance when personally controlled air movement is used by tropically acclimatized persons当热带适应的人使用个人控制的空气运动时,热舒适性,感知的空气质量和认知表现
Indoor Air
IF0
Synthesis of n-type semiconducting diamond film using diphosphorus pentaoxide as the doping source以五氧化二磷为掺杂源合成n型半导体金刚石膜

