EyaHaddad/tsl-ai-model

AI model development for Tunisian Sign Language recognition using MediaPipe landmarks and LSTM. # tsl-ai TensorFlow/TFLite hand-sign recognition workflow with MediaPipe real-time inference. ## Project Overview - `data_preprocessing.ipynb`: dataset preparation pipeline. - `lstm.ipynb`: baseline LSTM model training and export. - `lstm-cnn.ipynb`: hybrid LSTM-CNN architecture training and export. - `real_time_mediapipe.ipynb`: webcam inference with MediaPipe + TFLite. ## Environment Setup Prerequisites: - Python 3.10+ - `pip` Windows (PowerShell): ```powershell python -m venv .venv .\.venv\Scripts\Activate.ps1 pip install -r requirements.txt ``` macOS / Linux: ```bash python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt ``` ## Expected Workspace Structure ```text tsl-ai/ ├── data_preprocessing.ipynb # dataset preparation ├── lstm-cnn.ipynb # main TensorFlow training/export notebook ├── lstm.ipynb # baseline model experiments ├── real_time_mediapipe.ipynb # webcam inference pipeline ├── requirements.txt ├── artifacts/ │ ├── lstm_dataset.npz │ ├── lstm_dataset_meta.json │ └── tf_model_dir/ │ ├── baseline/ │ │ ├── model_float16.tflite │ │ └── model_float32.tflite │ └── lstmhybrid_tf/ │ ├── model_builtin_float32.tflite │ ├── model_float16.tflite │ └── model_float32.tflite └── mp/ └── hand_landmarker.task ``` ## Training and Export Flow Run `lstm-cnn.ipynb` end-to-end to: 1. Load dataset artifacts. 2. Expand sequences with frame duplication (target sequence length = 20). 3. Train the TensorFlow hybrid LSTM model. 4. Export Keras/SavedModel and TFLite variants. 5. Validate inference on generated TFLite output. ## Real-Time Inference Run `real_time_mediapipe.ipynb` to test webcam inference. - Baseline path uses `artifacts/tf_model_dir/baseline/model_float32.tflite`. - Hybrid path should prefer `artifacts/tf_model_dir/lstmhybrid_tf/model_builtin_float32.tflite`. ## Notes and Troubleshooting - Training notebooks …