purplegeminii/ECG_OCR

Fine-tuning Tesseract OCR for Ghana ECG postpaid electricity meter reading extraction — includes preprocessing, annotation, augmentation, training, evaluation, and a Flask inference API. # ECG Postpaid Meter Readings — Tesseract OCR Fine-Tuning Pipeline A robust, end-to-end pipeline for training/retraining a Tesseract OCR model specifically optimised for Ghana ECG (Electricity Company of Ghana) postpaid meter reading images. --- ## Table of Contents 1. Project Structure 2. Quick Start 3. Makefile Reference 4. Pipeline Phases 5. Configuration 6. Scripts Reference 7. Notebooks 8. Evaluation 9. Troubleshooting 10. System Overview --- ## Project Structure ``` ecg-ocr-project/ ├── Makefile # Convenient shortcuts for all pipeline operations (make help) ├── raw_images/ # Original meter reading photos ├── preprocessed/ # Cleaned/deskewed images ready for OCR ├── ground_truth/ # .gt.txt label files (paired with .tif) ├── augmented/ # Synthetically augmented images ├── eval_data/ # Holdout test set (auto-populated by 04_prepare_training_data.py) ├── corrections/ # Human-corrected samples for iterative retraining ├── results/ # Inference outputs, CSV reports, test_set.txt ├── models/ │ └── ecg_meter/ │ └── tessdata/ # Final .traineddata model installed here ├── tesstrain/ # git submodule — owns all training internals │ └── data/ │ └── ecg_meter-ground-truth/ # ← 04_prepare_training_data.py writes here ├── ecg-meter-api/ # Production REST API │ ├── app.py # Flask application │ ├── ocr.py # OCR engine wrapper │ └── model/ # Trained model (.traineddata) ├── scripts/ # All pipeline scripts ├── config/ # YAML configs ├── logs/ # Training logs, CER curves ├── notebooks/ │ └── ecg_ocr_analysis.ipynb # Interactive dataset analysis & preprocessing visualisation ├── tests/ # Unit tests └── docs/ # Additional documentation ``` --- ## Quick Start ```bash # 1. Clone the repo with the tesstrain submodule git clone --recurse-submodules cd ecg-ocr-proj …