American Sign Language Detection using EfficientNetB0

This project utilizes the EfficientNetB0 CNN architecture for image classification + detection of American Sign Language (ASL) alphabet images.

Setup and Installation

1. Install Required Dependencies

pip install -r requirements.txt
pip install kaggle

2. Download the ASL Alphabet Dataset

To download the dataset directly from Kaggle, ensure that your kaggle.json API key is properly set up. Follow these steps:

1. Move your kaggle.json file to the directory:

   • On Windows: C:/Users/{your_username}/.kaggle/
   • On macOS/Linux: ~/.kaggle/

2. Run the following command to download the dataset:

   kaggle datasets download -d debashishsau/aslamerican-sign-language-aplhabet-dataset

3. Unzip the dataset:

   unzip aslamerican-sign-language-aplhabet-dataset.zip

3. Train the Model

You can train the model locally using the dataset:

• Ensure you have at least 8GB of VRAM on a CUDA-supported GPU.
• Open the ASL_Alphabet_Classification.ipynb notebook and follow the instructions for training.

Note

Training may take around 1 hour to complete. After training, the model file efficientnet_model.pth will be saved under the models/ directory.

4. Running the Demo

Streamlit Demo (Image Classification):

cd demo
streamlit run main.py

Live Detection Demo:

python3 detect.py

Note

If you don't have a webcam, you can use the DroidCam app to turn your mobile phone into a webcam. Logs will be saved in the action_handler.log file.

5. Pre-trained Model

You can use the pre-trained model efficientnet_model.pth located in the models/ directory to perform inference on ASL images in your local environment or in Google Colab.

---

Feel free to report any issues you encounter.
Don't forget to ⭐ the repo!