Skip to content

Validating a Trained OCR Model

The valid command is used to evaluate a trained license plate OCR model on a labeled dataset, using the same configuration and preprocessing setup used during training.

Basic Usage

fast-plate-ocr valid \
  --model trained_models/2025-06-28_14-33-51/best.keras \
  --plate-config-file config/latin_plates.yaml \
  --annotations data/val.csv

Evaluate By Country

If your annotations include plate_region and your plate config defines plate_regions, you can evaluate metrics for each region:

fast-plate-ocr valid \
  --model trained_models/2025-06-28_14-33-51/best.keras \
  --plate-config-file config/latin_plates.yaml \
  --annotations data/val.csv \
  --evaluate-by-region

To save per-region metrics to a JSON file instead of printing them, use --region-metrics-output.

Output

  • Evaluation metrics will be printed to the terminal (e.g., accuracy, loss).
  • The script automatically compiles the model using the metrics defined during training.
  • It does not save new weights or modify the model.
  • If the model includes a region head, region metrics like region_acc, region_top3_acc, and region_macro_f1 will also appear.
  • With --evaluate-by-region, a JSON summary with per-region metrics is printed after the global evaluation.
  • With --region-metrics-output, per-region metrics are saved to a JSON file instead of being printed.

Example Output

40/40 ━━━━━━━━━━━━━━━━━━━━ 10s 100ms/step - loss: 0.2185 - plate_acc: 0.921 - cat_acc: 0.983 - top_3_k: 0.998