MambaPose: A Human Pose Estimation Based on Gated Feedforward Network and Mamba
2024

MambaPose: A Human Pose Estimation Method

publication Evidence: high

Author Information

Author(s): Zhang Jianqiang, Hou Jing, He Qiusheng, Yuan Zhengwei, Xue Hao

Primary Institution: Taiyuan University of Science and Technology

Hypothesis

Can the Mamba-based human pose estimation method improve accuracy in detecting keypoints in complex scenes?

Conclusion

The proposed MambaPose method significantly improves the accuracy of human pose estimation, especially in dense crowds and for small targets.

Supporting Evidence

  • The MambaPose algorithm improved AP50 by 1.1% compared to typical algorithms.
  • Experimental results showed that the proposed method effectively detects keypoints in complex scenes.
  • The model outperformed existing methods in both subjective and objective evaluations.

Takeaway

This study created a new way to find where people's body parts are in pictures, making it easier to see them even when there are many people close together.

Methodology

The study used a GMamba structure for feature extraction, slice downsampling for resolution reduction, and an adaptive threshold focus loss to enhance keypoint detection.

Limitations

The model may face challenges in real-time performance and computational resource limitations due to a large number of parameters.

Digital Object Identifier (DOI)

10.3390/s24248158

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication