AbayoHJovin/rwanda-water-meter-dataset

# Rwanda Water Meter Reading System Automatic water meter reading using a **two-model YOLOv8 pipeline**. A digit-detection model reads the digits; an OBB (oriented bounding box) model corrects for meter tilt before the digit model runs. ``` Photo → OBB model → Deskew → Digit model → Reading ``` --- ## Project structure ``` rwanda-water-meter-reading-system/ │ ├── 01_label_images.py GUI labeler — draw bounding boxes for digits, meter, window ├── label_window.py OBB labeler — draw rotated window box + capture tilt angle │ ├── 02_prepare_dataset.py Build train/val/test dataset from raw labeled images ├── 03_train.py Train the digit-detection model (YOLOv8-detect) ├── 04_predict.py Batch inference on the test set ├── 05_retrain.py Continue training from the latest checkpoint ├── 06_prepare_obb_dataset.py Build OBB train/val dataset from rotated-window images ├── 07_train_obb.py Train the window-rotation model (YOLOv8-OBB) ├── predict_and_read.py Single-image or folder inference with annotated output │ ├── WaterMeter_Pipeline_v3.ipynb Colab notebook — calls the scripts above ├── requirements.txt └── README.md ``` --- ## Quickstart ### 1. Install dependencies ```bash pip install -r requirements.txt ``` Or manually: ```bash pip install ultralytics opencv-python PySide6 ``` ### 2. Label images **Digit labels** (axis-aligned boxes): ```bash python3 01_label_images.py ``` Classes: `meter`, `window`, `0`–`9`, `unknown`. **OBB labels** (rotated window + tilt angle): ```bash python3 label_window.py ``` | Key | Action | |-----|--------| | `W` | Draw window OBB | | `M` | Draw meter box | | `0`–`9` | Draw digit box | | `U` | Mark unknown digit | | `N` / `P` | Next / previous image | | `S` / `Ctrl+S` | Save labels | | `Delete` | Remove selected box | | `Ctrl+Z` | Undo | | `Shift+Scroll` | Fine rotation | | `Ctrl+Left/Right` | Rotate ±15° | | `Ctrl+R` | Enter custo …