This repository contains the code and resources for a custom hand pose detection model trained using the YOLOv8n-pose framework by ultralytics. The model is trained on a custom dataset of hand keypoints available on Kaggle.
Important Note: Due to computational limitations, the model has only been trained to 50% of its potential. However, it still demonstrates excellent performance.
Project Goal
The primary objective of this project is to contribute to the development of state-of-the-art (SOTA) models for hand pose detection, specifically targeting applications in sign language classification and AR/VR.
What's Included:
- Training scripts for the YOLOv8n-pose model on the custom hand keypoint dataset.
cd ./train
- Pre-trained model weights (50% trained).
cd ./model
- Configuration files and scripts for inference with the trained model.
cd ./inference
Future Developments:
- Continue training the model to achieve full capacity and improve accuracy.
- Explore advanced techniques for hand pose estimation and landmark detection.
- Integrate the model into sign language classification and AR/VR projects.
Contribution
We welcome contributions from the community to enhance this project. Feel free to:
- Fork the repository and experiment with different training configurations.
- Implement additional functionalities for hand pose estimation.
- Share your ideas for integrating the model into sign language and AR/VR applications.
Let's build better hand pose detection models together!
GitHub @RionDsilvaCS · Linkedin @Rion Dsilva · Twitter @Rion_Dsilva_CS