Ysen1903/LSM_Darija_Backend

An end-to-end framework for Moroccan Sign Language (LSM) recognition using Mediapipe for keypoint extraction and LSTM (Long Short-Term Memory) neural networks for sequence classification # LSM Darija - Sign Language Recognition System An end-to-end framework for Moroccan Sign Language (LSM) recognition using Mediapipe for keypoint extraction and LSTM (Long Short-Term Memory) neural networks for sequence classification. --- ## Overview This project provides a complete pipeline to collect, train, and deploy a real-time sign language recognition system tailored for Moroccan Darija. It leverages high-resolution hand and body landmarks to understand temporal gesture sequences. ### Key Features - **Real-time Inference**: Interactive webcam interface with visual feedback and confidence scoring. - **Automated Data Collection**: Streamlined recording process for building custom gesture datasets. - **Deep Learning Pipeline**: PyTorch implementation of an LSTM-based architecture for sequential data. - **Visual Feedback**: UX-enhanced displays during both collection and prediction phases. --- ## Project Structure ```text ├── data_collection/ # Tools for recording gestures and extracting keypoints │ ├── main_data.py # Main script for automated data gathering │ └── utils/ # Helpers for folder initialization and keypoint extraction ├── src/ # Core library │ ├── LSTM_models/ # Model architectures (LSTM, etc.) │ ├── utils/ # Mediapipe handlers and UI visualization tools │ └── config.py # Global settings (actions, sequence length, paths) ├── train_model/ # Model training environment │ ├── train.py # Training script │ ├── model.py # Model definition for training │ └── dataset.py # PyTorch Dataset for loading .npy keypoints ├── main_inference.py # Main application for real-time recognition ├── requirements.txt # Project dependencies └── models/ # (Created upon training) Storage for .pth weight files ``` --- ## Installation 1. **Clone the repository:** ```bash git clone github.com cd LSM_Darija_Backend ``` 2. ** …