💃 2D and 3D Dance Pose Estimation

• 2D Dance pose Estimation using Yolov7

• 3D Dance pose Estimation using Mediapipe

Abstract

This project detect the 2D and 3D dance pose data using Yolov7 and Mediapipe, respectively. A test video has been given in the format of google file id. Huamn pose data will be saved in the format of json and will be saved in the output folder.

If need to change another video, delete input folder videos and put new video under input folder. Note could solve multiple videos.

Getting Started

Clone the repository.

git clone https://github.com/Zhijie-He/2D-Dance-Pose-Estimation-with-YOLOv7.git

Go to the cloned folder.

cd 2D-Dance-Pose-Estimation-with-YOLOv7

Steps to run Code

Create virtual environment

### For Linux Users
python3 -m venv yolov7_dance
source yolov7_dance/bin/activate

### For Window Users
python3 -m venv yolov7_dance
cd yolov7_dance
cd Scripts
activate
cd ..
cd ..

Installation

Install requirements with mentioned command below.

pip install -r requirements.txt

Object detection and huamn pose estimation

In the configuration file (config/cfg.yaml), a test video has been gived which can be downloaded by gdown.

Run the code

python main.py --device [gpu/cpu] --dim [2D/3D]

• device DEVICE cpu/0,1,2,3(gpu), default gpu
• dim {2D,3D} choose to extract 2D or 3D huamn pose data, default 3D

The huamn pose estimation data will be saved in output file with the json format.

Plot human pose with json file using matplotlib

Reference

• https://github.com/WongKinYiu/yolov7
• https://github.com/SkalskiP/sport
• https://github.com/google/mediapipe