Monocular Estimation of Body Orientation In the Wild


Body orientation estimation provides crucial visual cues in many applications, including robotics and autonomous driving. It is particularly desirable when 3D pose estimation is difficult to infer due to poor image resolution, occlusion, or indistinguishable body parts.

We present COCO-MEBOW (Monocular Estimation of Body Orientation in the Wild), a new large-scale dataset for orientation estimation from a single in-the-wild image. The body-orientation labels for 133380 human bodies within 55K images from the COCO dataset have been collected using an efficient and high-precision annotation pipeline. There are 127844 human instance in training set and 5536 human instance in validation set.

Based on MEBOW, we established a simple baseline model for human body orientation estimation. The code and trained models are available on Github.


Image of MEBOW


MEBOW dataset is only for research purposes. Commercial use is not allowed.

Images of MEBOW all come from COCO dataset. Please download images from Click “2017 Train images [118K/18GB]” and “2017 Val images [5K/1GB]” in the COCO download page. Please email to get access to human body orientation annotation. You will usually get a reply within 24 hours (no more than 72 hours).


If you use our code or models in your research, please cite with:

  title={MEBOW: Monocular Estimation of Body Orientation In the Wild},
  author={Wu, Chenyan and Chen, Yukun and Luo, Jiajia and Su, Che-Chun and Dawane, Anuja and Hanzra, Bikramjot and Deng, Zhuo and Liu, Bilan and Wang, James Z and Kuo, Cheng-hao},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},