It is a model that used to predict American Sign Language that contains 26 alphabets. The model is trained using a single GPU.
This project was built upon python 3.12.2.
- You could create the virtual environment by using anaconda command
conda create -n rps python=3.12- Install the requirements
pip install -r requirements.txt- Download the given pretrained model below and put those under
pretrainedfolder - execute the inference using your webcam
cd AmericanSignLanguage
python -m asl_detection.infer_webcamThe methods used are YOLOV8 with the following results:
| model | size (pixels) | mAP50 val | mAP50-95 val | FLOPs (B) | Link |
|---|---|---|---|---|---|
| YOLOv8n | 640 | 0.953 | 0.794 | 8.7 | download |
| YOLOv8x | 640 | 0.96 | 0.819 | 257.8 | download |
Note : I trained with 2 backbone options that you could choose, the smallest one (YOLOv8n) or the biggest one (YOLOv8X).
If any feature requests please write on this github issues
Hopefully this pretrained model benefits for those who need it. If you are one that benefits this project, please give me a high five by assigning a STAR.