cylab-africa/fidel_dataset_benchmarks

# Fidel Baseline: CRNN-based Amharic OCR Training & Evaluation A complete end-to-end pipeline for training and evaluating a CRNN-based Amharic OCR model. Features: * 🔧 **Configurable** via YAML * 🐍 **Installable** as a module (and editable during development) * 🎛️ **Weights & Biases** integration for experiment tracking * ☁️ Works with Conda or plain `venv` + `pip` --- ## 📋 Table of Contents 1. Prerequisites 2. Installation 3. Configuration * .env * config YAML 4. Project Layout 5. Usage * As a module * Via console-script (optional) 6. Conda-only Quick-Start 7. Scripts Reference 8. TrOCR variation 9. Notes --- ## 🛠️ Prerequisites * **OS:** Linux or macOS (Unix-like) * **Python:** 3.8+ * **Git** (for cloning) * **(Optional)** CUDA-enabled GPU for faster training * **Weights & Biases** account (for experiment logging) --- ## ⚙️ Installation 1. **Clone the repo** ```bash git clone github.com cd fidel_dataset_benchmarks ``` 2. **Install in editable mode** ```bash # using pip+venv or inside any Python env: python3 -m pip install --upgrade pip python3 -m pip install -e . ``` This does two things: * Puts `fidel_baseline` (and other packages under `src/`) on your PYTHONPATH * (If configured) creates a `fidel-benchmark` console script --- ## 🔧 Configuration ### .env file In the project root, create a `.env` (and add it to `.gitignore`): ```bash # .env WANDB_API_KEY= ``` ### config YAML All hyperparams & paths live in a YAML. E.g. `src/configs/crnn_config.yaml`: Adjust these values and file paths to your setup. --- ## 📁 Project Structure ``` fidel_dataset_benchmarks/ ├── .env ├── pyproject.toml # installable package config ├── requirements.txt # for non-editable install ├── src/ │ ├── configs/ │ │ ├── crnn_config.yaml │ │ └── main.yaml │ ├── fidel_baseline/ # python package │ │ ├── __init__.py │ │ └── benchmark.py │ ├── my_datasets/ │ ├── models/ …