Skip to main content
ALPR Vue relies on two ONNX models that run locally in your browser — no cloud API, no remote inference. The first model detects where license plates are in the image; the second reads the characters on each plate. Both models load directly in your browser tab using ONNX Runtime Web.
The plate detector scans each frame and draws bounding boxes around any license plates it finds.
PropertyValue
Model fileyolo-v9-t-384-license-plates-end2end.onnx
ArchitectureYOLOv9 (single-pass object detection)
Input resolution384 × 384 pixels
Sourceopen-image-models
How it works: The model resizes each frame to 384 × 384 pixels, then analyzes the entire image in a single forward pass through the neural network. It divides the image into a grid and predicts bounding boxes and confidence scores for each cell. The result is a set of bounding boxes — one for each plate region the model found.The model is a compact YOLOv9-t variant, designed to run efficiently on resource-limited devices like phones and laptops. It has been specifically trained on European plates under a variety of lighting conditions and camera angles.

​
Model format

Both models are stored in ONNX (Open Neural Network Exchange) format. ONNX is an open standard for representing machine learning models, which makes it possible to run models trained in Python directly in the browser. ALPR Vue uses ONNX Runtime Web to run inference. This library compiles the models to WebAssembly, which runs natively in modern browsers without any plugins or extensions. The result is near-native inference speed entirely within your browser tab.

​
Limitations

The models are optimized for European license plates. Detection and OCR accuracy may be lower for plates from other regions, especially those that use different character sets, plate dimensions, or layouts.
  • Regional coverage: Both models were trained primarily on European plates. Accuracy varies for plates from North America, Asia, and other regions.
  • Device performance: Inference speed depends on your device’s CPU. Older or low-power devices may experience slower frame rates.
  • Browser requirements: ONNX Runtime Web requires WebAssembly support. All modern browsers support this, but very old browser versions may not.
The models are downloaded once on your first visit and cached by your browser. On subsequent visits, the app loads them from the local browser cache — no network request is needed, and the app works fully offline.