Official code of 《The Devil is in the Details: Delving into Unbiased Data Processing for Human Pose Estimation》
Official code of 《AID: Pushing the Performance Boundary of Human Pose Estimation with Information Dropping Augmentation》
News
[2021/1/12] A new version of UDP paper is provided with more clear and more detailed methodology explaination, extra experimental results, and more discoveries. ArXiv.
[2020/11/23] UDP for mmpose is provided in HuangJunJie2017/mmpose alone with pretrained models in BaiduDisk(dsa9). Examples for both top-down paradigm and bottom-up paradigm are provided in this branch.
[2020/11/04] We propose UDPv1 with LOSS.KPD=3.5. The performance of UDPv1 is superior when compared with UDP in coco dataset.
[2020/10/26] We get a better tradeoff between speed and precision by applying UDP to the state-of-the-art Bottom-Up methods.
[2020/8/23] We win the 2020 COCO Keypoint Detection Challenge with UDP!
[2020/6/12] UDP for hrnet and UDP for RSN are provided.
Data preparation
For coco, we provide the human detection result and pretrained model at BaiduDisk(dsa9)
Citation
If you use our code or models in your research, please cite with:
@InProceedings{Huang_2020_CVPR,
author = {Huang, Junjie and Zhu, Zheng and Guo, Feng and Huang, Guan},
title = {The Devil Is in the Details: Delving Into Unbiased Data Processing for Human Pose Estimation},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}
@article{huang2020aid,
title={AID: Pushing the Performance Boundary of Human Pose Estimation with Information Dropping Augmentation,
author={Huang, Junjie and Zhu, Zheng and Huang, Guan and Du, Dalong},
journal={arXiv preprint arXiv:2008.07139},
year={2020}
}