PoseGate-Former: Transformer Encoder with Trainable Gate for 3D Human Pose Estimation Using Weakly Supervised Learning
- Publisher:
- Springer International Publishing
- Publication Type:
- Chapter
- Citation:
- Neural Information Processing, 2021, 1517 CCIS, pp. 266-274
- Issue Date:
- 2021-01-01
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Weakly supervised learning for 3D human pose estimation can learn a real human structure, but it generally has lower accuracy on reconstructing 3D poses. In this work, we present a 3D pose estimation model using a Transformer encoder based architecture with a trainable gate, PoseGate-Former. The model is trained using individual images from a weakly supervised learning approach. It can reduce possibility of overfitting on some action categories due to the addition of a trainable gate to the Transformer encoder. We evaluated this model on two benchmark datasets: Human3.6M and HumanEva-I. The experimental results show that this model can obtain substantially better accuracy in all action categories of 3D human poses in the datasets compared with some fully-supervised 3D pose estimation approaches.
Please use this identifier to cite or link to this item: