AI FinessTrainer

AI-FitnessTrainer using YOLOv8-pose ?️

In today's world, a healthy lifestyle is becoming increasingly relevant, and with it, interest in sports activities is growing. However, gaining experience and knowledge in this field can be a challenging task for many people. In this context, the application of artificial intelligence (AI) in the sports sector becomes a key element of successful training, analysis, and development of sports teams and individual athletes.

This repository provides a set of tools to help you improve your technique for the following exercises: front squats, wide-arm push-ups, biceps push-ups, reverse push-ups. This intelligent assistant analyzes your technique in real time, evaluates your posture using an AI model (yolov8-pose) and gives you feedback on your form.

A counter for correctly completed sets and so-called attempts to perform the exercise correctly has also been added. This will help you better understand your exercise statistics.

About the pose estimation model and the dataset used

This project uses a trained YOLOv8m-pose model. However, you can use the weights of 2 other trained models: YOLOv8n-pose and YOLOv8s-pose located in the models/yolo and models/yolo2 folders respectively, but these results are worse than YOLOv8m-pose.

These models are pose detection models that are trained on COCO-pose typed data. This dataset includes 17 keypoints. For clarity, below is a markup image.

More information on the dataset can be found here: COCO-Human-Pose and Ultralytics: COCO-Pose Dataset.

How to use

1. Clone repository.

git clone https://github.com/KKopilka/AI-FinessTrainer.git

2. Install the requirements.

pip install -r requirements.txt

3. Run the script.

python manual.py 

4. It is possible to run the project with streamlit.

streamlit run app/live.py

5. If you want to run the project through docker. Documentation  Deploy Streamlit using Docker.

docker build -t streamlit .

docker-compose up -d

Project roadmap

• Train a model for human pose estimation.

• Integration of the model into the project, processing of key points.

• Add exercises for major muscle groups.

• Add a counter for approaches and attempts.

• Run locally or through a browser (streamlit).

• Launching via Docker.

Some ideas

This project is not a fully finished version, so it can still be finalized.

Here are some ideas on how to improve this project are as follows:

• Add more exercises.

• Add more statistics to the program.

• Add a web/mobile app.

• Add sound accompaniment.

• Convert the project to an .exe file.