Sign Language Gesture Classifier

A deep learning project that classifies American Sign Language (ASL) gestures from images using a trained ResNet model. This project demonstrates image classification, transfer learning, and interactive inference using FastAI and Gradio.

Project Overview

This project aims to help in recognizing ASL gestures to facilitate communication between hearing and non-hearing individuals. Key features:

• Classifies 26 ASL alphabet gestures
• Trained on a custom ASL dataset
• Supports image input for real-time predictions
• Easy-to-use Gradio interface for demonstration

Technologies Used

• Python 3.x
• FastAI
• PyTorch
• Gradio [optional, for interactive demo(https://8e784fcf3c0e85d22a.gradio.live/)]
• Jupyter Notebook / Google Colab

Project Structure

SignLanguageClassifier/
│
├── export2.pkl            # Trained model weights (large file, download separately)
├── SignLanguage.ipynb    # Jupyter notebook with training & inference code
├── README.md             # Project description
└── requirements.txt      # Python dependencies

Setup Instructions

1. Clone the repository

git clone https://github.com/your-username/SignLanguageClassifier.git
cd SignLanguageClassifier

2. Install dependencies

pip install -r requirements.txt

3. Download the trained model

• Model file (export2.pkl) is hosted on Hugging Face
• Save it in the project directory.

4. Run inference

• Open the notebook SignLanguage.ipynb
• Use the provided code cells to test predictions on your images.

---

Usage Example

from fastai.vision.all import load_learner, PILImage

# Load model
learn = load_learner('export2.pkl')

# Predict gesture
img = PILImage.create('sample_image.jpg')
pred, pred_idx, probs = learn.predict(img)

print(f"Predicted gesture: {pred}; Probability: {probs[pred_idx]:.04f}")

contributing

Contributions, suggestions, and improvements are welcome! Please fork the repo, create a branch, and submit a pull request.

---