README.md

---
title: AquaVision AI Water Meter Reader
emoji: 💧
colorFrom: blue
colorTo: green
sdk: static
pinned: false
short_description: Privacy-first AI water meter reading in your browser
---

# AquaVision AI Water Meter Reading System

This free Static Space runs the exported YOLO model directly in the visitor's
browser with ONNX Runtime Web. The uploaded photograph is processed locally
and is not sent to an inference server.

The interface includes final-model validation metrics and evidence from the
52-image combat test. During prediction, low-confidence digit detections are
shown as `?` in the safe reading while preserving the model's raw digit guess;
`unknown` is post-processing logic, not a trained class in the final model.
The browser decoder can also recover one very-low-confidence trailing digit
when it is inside the detected window and follows the established digit spacing.
Recovered digits remain unknown (`?`) unless they pass the unknown threshold.

The interface is organized into Reader, Performance, Settings Guide, History,
and Team tabs. The guide explains detection confidence, IoU overlap filtering,
and the post-processing unknown threshold, and provides a one-click recommended
10% / 45% / 40% preset. The Performance tab contains the four validation
metrics and five full-resolution evaluation plots without requiring users to
scroll through them from the reader workspace.

The Reader tab is viewport-first: its introduction and upload target are
compact, and the primary Run Analysis action appears immediately after upload
so the upload-to-predict workflow is visible without scrolling on common
desktop displays.

All three detection settings are displayed together in an always-visible,
compact strip above the upload workspace. No drawer or hidden settings action
is required. A Team tab credits the three Y3 project members: Benite, Beni
Samuel, and Kagabo Lucky.

The Reader uses one locally hosted AquaVision YOLO12n ONNX model at `960 x 960`.
The Performance tab contains the validation evidence for this same model.

Source model: [`kagabo/hf_water_meter_models`](https://huggingface.co/kagabo/hf_water_meter_models)