Multi-person pose estimation (aka, keypoint detection) is one of the fundamental computer vision tasks and has a wide range of applications such as action recognition, augmented reality, human computer interaction, pedestrian tracking and re-identification, etc.
YOLOv8 pretrained Pose models, Detect, Segment and Pose models are pretrained on the COCO dataset, while Classify models are pretrained on the ImageNet dataset.
The command bellow will do a pose estimation on a default image
cd yolov8
python pose_estimation_yolov8.py
The command bellow will do a pose estimation on a given video path
cd yolov8
python pose_estimation_yolov8.py --input "../media/videos/people-walk.mp4"
PyTorch provides a pre-trained Keypoint RCNN model with ResNet50 base which has been trained on the COCO keypoint dataset. Keypoint RCNN is an algorithm for finding keypoints on images containing a human.
cd pytorch
python pose_estimation_pytorch.py
In progress ...