A simple benchmark on current state-of-the-art pose estimation methods.
Current methods:
- AlphaPose (MXNet)
- Alphapose (PyTorch) (comming soon)
- Detectron2 Coco Keypoints
- HRNet: High-Resolution Network
- Higher HRNet (comming soon)
- Simple Baselines for Human Pose Estimation (MXNet)
- Simple Baselines for Human Pose Estimation (PyTorch) (comming soon)
The following results were obtained using high quality videos (1920x1080) with duration between 2 and 10 minutes. Due to nature of the videos, we cannot put them publicly available.
| Method | Repo | FPS |
|---|---|---|
| Alphapose (MXNET) | https://gluon-cv.mxnet.io/model_zoo/pose.html | 6.68 |
| Alphapose (PyTorch) | https://github.com/MVIG-SJTU/AlphaPose | *** |
| Detectron2 Pose Estimator | https://github.com/facebookresearch/detectron2 | 7.96 |
| Higher HRNet | https://github.com/HRNet/HigherHRNet-Human-Pose-Estimation | |
| HRNet | https://github.com/stefanopini/simple-HRNet | 11.16 |
| OpenPose | https://github.com/CMU-Perceptual-Computing-Lab/openpose | |
| Simple Baseline for Human Pose Estimation (MXNET) | https://gluon-cv.mxnet.io/model_zoo/pose.html | 5.28 |
| Simple Baseline for Human Pose Estimation (PyTorch) | https://github.com/microsoft/human-pose-estimation.pytorch |