1chimaruGin / Pose-Estimation

All body, body and hand pose estimation with faster inference.

Faster Pose Estimation

WHOLE BODY, BODY AND HAND POSE ESTIMATION with YOLOv5 and MMDetection

Requirements

• mmcv >= 1.4.0
• mmpose >= 0.26.0

Install MMCV and MMPose according to documentation

Other requirements can be installed with pip install -r requirements.txt.

Clone the repository recursively:

$ git clone --recursive [e-mail]:1chimaruGin/Pose-Estimation.git

Then download a YOLO model's weights,

Hand pose estimation - hand

All body and body estimation - body

Place them in weights/.

Pose Esitmation

Change the config acording to your needs.

DETMODEL:
  CHECKPOINT            : weights/yolov5s6.pt
  HALF                  : False
  CLASSES               : 0
  IMGSZ                 : 1280
  CONF_THRESH           : 0.5
  IOU_THRESH            : 0.5

POSEMODEL:
  CONFIG                : configs/body/2d_kpt_sview_rgb_img/topdown_heatmap/coco/hrnet_w48_coco_384x288.py
  CHECKPOINT            : https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_384x288-314c8528_20200708.pth

$ python pose.py --source /path/to/video.mp4

References

• https://github.com/ultralytics/yolov5
• https://github.com/open-mmlab/mmpose

Citations

{
  @misc{mmpose2020,
    title={OpenMMLab Pose Estimation Toolbox and Benchmark},
    author={MMPose Contributors},
    howpublished = {\url{https://github.com/open-mmlab/mmpose}},
    year={2020}
}