Skip to content

Visualize Predictions

Once your model is trained, you can visually inspect how it performs on unseen images using:

fast-plate-ocr visualize-predictions \
  --model arg_cnn_ocr.keras \
  --img-dir benchmark/imgs \
  --plate-config-file arg_cnn_ocr_config.yaml

What You'll See

For each image, the model will:

  • Predict the license plate text
  • Overlay each predicted character with its confidence score
  • Optionally color low-confidence characters (default: red if below 0.35)

Visualize Predictions

Example: Filter Uncertain Plates

fast-plate-ocr visualize-predictions \
  --model model.keras \
  --img-dir raw_eval_imgs \
  --plate-config-file config.yaml \
  --filter-conf 0.5

This filters out predictions unless at least one character has confidence < 0.5, helpful when manually reviewing low-quality predictions.