CoolFace
Modelpublic

carlravel/indonesian-plate-recognizer-yolo11

sourceHugging Facemitupdated 1mo agoView on Hugging Face
0likes35downloads
Model Card

🇮🇩 Indonesian License Plate Character Recognizer (YOLO11)

A fine-tuned YOLO11 Nano character recognition (OCR) model trained to detect and classify 36 alphanumeric character classes (0-9, A-Z) on cropped Indonesian vehicle license plates.

Part of a 2-stage Automatic License Plate Recognition (ALPR) system:

  1. 1.Stage 1 ([Plate Detector](https://huggingface.co/carlravel/indonesian-plate-detector-yolo11)): Localizes the license plate bounding box.
  2. 2.Stage 2 (This Model): Detects and transcribes individual alphanumeric characters.

📊 Benchmark & Performance

Evaluated on the validation split of the Indonesian License Plate Recognition Dataset:

MetricScore
Precision96.24% (0.9624)
Recall96.34% (0.9634)
mAP@0.598.40% (0.9840)
mAP@0.5:0.9572.01% (0.7201)
GPU Inference Latency2.41 ms (~415 FPS)
Preprocess Latency0.81 ms
Postprocess Latency1.32 ms

🖼️ Sample Visual Detections & OCR Reconstruction

[image]

Training Loss Curves & Normalized Confusion Matrix

Training Metrics ProgressionNormalized Confusion Matrix
[image][image]

🛠️ Usage with Ultralytics

Quickstart (Python)

python
from ultralytics import YOLO

# Load character recognizer
model = YOLO("plate_recognizer_yolo11n.pt")

# Run inference on cropped plate image
results = model.predict("cropped_plate.jpg", conf=0.35)

# Display bounding boxes
for r in results:
    r.show()

📦 Export Formats Included

  • —plate_recognizer_yolo11n.pt: PyTorch weights.
  • —plate_recognizer_yolo11n.onnx: Optimized ONNX runtime export.

📄 License & Attribution

This model and its weights are open-source and released under the [MIT License](LICENSE).