vendor/ViTPose/docs/en/papers/algorithms/videopose3d.md

# 3D human pose estimation in video with temporal convolutions and semi-supervised training

<!-- [ALGORITHM] -->

<details>
<summary align="right"><a href="http://openaccess.thecvf.com/content_CVPR_2019/html/Pavllo_3D_Human_Pose_Estimation_in_Video_With_Temporal_Convolutions_and_CVPR_2019_paper.html">VideoPose3D (CVPR'2019)</a></summary>

```bibtex
@inproceedings{pavllo20193d,
  title={3d human pose estimation in video with temporal convolutions and semi-supervised training},
  author={Pavllo, Dario and Feichtenhofer, Christoph and Grangier, David and Auli, Michael},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={7753--7762},
  year={2019}
}
```

</details>

## Abstract

<!-- [ABSTRACT] -->

In this work, we demonstrate that 3D poses in video can be effectively estimated with a fully convolutional model based on dilated temporal convolutions over 2D keypoints. We also introduce back-projection, a simple and effective semi-supervised training method that leverages unlabeled video data. We start with predicted 2D keypoints for unlabeled video, then estimate 3D poses and finally back-project to the input 2D keypoints. In the supervised setting, our fully-convolutional model outperforms the previous best result from the literature by 6 mm mean per-joint position error on Human3.6M, corresponding to an error reduction of 11%, and the model also shows significant improvements on HumanEva-I. Moreover, experiments with back-projection show that it comfortably outperforms previous state-of-the-art results in semi-supervised settings where labeled data is scarce.

<!-- [IMAGE] -->

<div align=center>
<img src="https://user-images.githubusercontent.com/15977946/146525184-34a14153-5cab-4fe0-8500-018c99a5f647.png">
</div>