Pose estimation

code yolov7-w6-pose.pt

See keypoint.ipynb.

Installation and Environment setup

Install

git clone https://github.com/YunghuiHsu/yolov7.git
cd yolov7/

# Install the dependency package
pip3 install -r requirements.txt
# Please decide if you want to comment out opencv-python, torch, torchvision depending on the platform you are using

• [optional] download weight

# yolov7-w6-pose.pt
wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-w6-pose.pt
# yolov7-tiny.pt
wget wget https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-tiny.pt

• [optional] Switching branches

# git checkout -b detect_pose origin/detect_pose # default. can run detect or pose estimation tasks directly
git checkout -b log_metric origin/log_metric # for performance testing
git checkout -b deploy_ui origin/deploy_ui  # for deploying GUI in Jetson with YOLOv7() class

Update

• brach:deploy_ui

  • Refactor and encapsulate as Yolov7 class
    • yolov7/YOLOv7_Engine.py
  • Add the simple UI example file implemented by pyQT
    • LiveFrameThread.py
    • pyqtUI_Main.py

  sudo apt install python3-pyqt5
  # sudo apt install pyqt5-dev-tools  
  
  python3 pyqtUI_Main.py
  

• brach:log_metric

  • Add the function of logging performance during inference()

    When running the python command in the terminal, add --log_exp and it will start logging the values of inference time, FPS, Latency and hyperparameters, which requires the libraries for wandb and mlflow to be installed.

• brach:detect_pose

  • Add support for "pose estimation" from the official YOLOv7 repository

    Load yolov7-w6-pose.pt weights directly into the Pytorch environment for humane skeleton detection(pose estimation)

    detect_pose.py modified from detect.py, adding support for "pose estimation" to the original detect.py architecture, without modifying, adding any utils files. Directly download and ready to use

    For environment setup and package installation, please refer to the official YOLOv7 repository

    從YOLOv7官方倉庫中增加對 "人體姿勢估計 "的支持，可以簡單地將yolov7-w6-pos.pt權重直接加載到Pytorch環境中進行人體姿態偵測

    detect_pose.py修改自detect.py，在原detect.py架構中加入了對 "姿勢估計 "的支持，沒有修改、添加任何utils文件。直接下載並可直接使用

Inference for Pose estimation

• On the video:

python detect_pose.py --weights yolov7-w6-pose.pt --conf 0.05 --iou-thres 0.65 --img-size 1280 --source yourvideo.mp4 --no-trace 

• On the image:

python detect_pose.py --weights yolov7-w6-pose.pt --conf 0.05 --iou-thres 0.65 --img-size 1280 --source inference/images/horses.jpg --no-trace 

• On the webcam:
  add 0 after --source

python detect_pose.py --weights yolov7-w6-pose.pt --conf 0.05 --iou-thres 0.65 --img-size 1280 --source 0 --no-trace 

Test

• Crowded multi-person scenes

• Upside-down human posture

Citation

@article{wang2022yolov7,
  title={{YOLOv7}: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors},
  author={Wang, Chien-Yao and Bochkovskiy, Alexey and Liao, Hong-Yuan Mark},
  journal={arXiv preprint arXiv:2207.02696},
  year={2022}
}