rayen-feb/Tunisian-Sign-Language-recognition

Deep Learning-based system for Tunisian Sign Language recognition using CNN and Transfer Learning. Supports image classification and real-time detection. # 🤟 Tunisian Sign Language Recognition A deep learning project using **MobileNetV2** to recognize Tunisian Sign Language gestures from images. The system integrates **MediaPipe** for hand detection, supports ~57 classes, and provides predictions as **text, confidence scores, and Arabic voice spelling** via gTTS. --- ## 📸 Preview --- ## 🚀 Quick Start 1. Install dependencies: ```bash pip install -r requirements.txt Train the model: bash python main.py Test on a single image: bash python predict.py path/to/image.jpg Launch the Gradio web app: bash python app_gradio.py Then open the local URL (default: 127.0.0.1). 🌐 Web App (Gradio) Upload, drag & drop, or capture a photo of a sign to get predictions with text + confidence scores. Features 🖐️ Two modes: Single Image → predict one sign Sentence Prediction → combine multiple signs 📷 Webcam capture with a 3‑second countdown ✋ Hand cropping via MediaPipe 🔊 Voice spelling in Arabic (gTTS) 📊 Top‑3 predictions with confidence bars 🖼️ Example images from Data/test/ 🧠 Hand Model hand_crop.py uses MediaPipe HandLandmarker. If hand_landmarker.task is missing, the app falls back to a centered crop (less precise but functional). 📚 Classes Auto‑detected from Data/raw/ folders — currently ~57 signs. 📊 Status See [Il semble que le résultat n’était pas sûr à afficher. Changeons un peu et essayons autre chose !] for pending tasks and improvements. 🔗 Links 📂 GitHub Repository 🎥 Live Demo (local): 127.0.0.1 (Deploy on Hugging Face Spaces or Streamlit Cloud for public access) 🔮 Future Work Expand dataset with more Tunisian sign classes Deploy on Hugging Face Spaces for public demo Add real‑time video recognition with continuous prediction Improve accuracy with transfer learning and fine‑tuning ---