This repository contains the application for the visualization of human pose tracking and estimation. The application can be used to run pose estimation on a video with enabled or disabled pose tracking and save the visualized video along with the results in a JSON file. The JSON files can then be opened in the application and visualized on the video.
It is written in Python 3.10 and uses packages MMDet, MMTrack and MMPose. These packages use PyTorch and CUDA supported graphics card is needed to run the application. The user interface is made using the PyQt5 package.
- Python 3.10
- PyQt5
- PyTorch 1.13.0
- MMDet 2.28.2
- MMTrack 0.14.0
- MMPose 0.29.0
For installation, check Installation.md.
Pre-trained checkpoints are available for download on Google Drive. Extract the checkpoints to the main repository folder and make sure the checkpoints are located in PoseEstimationApp/Checkpoints.
These checkpoints were not trained by me, they were downloaded from the MMDet, MMTrack and MMPose repositories where they are available to download for free.