Metrics¶

The following metrics are tracked during model training and validation to help evaluate OCR model performance at both character level and plate level granularity.

Available Metrics¶

During training, you will see the following metrics:

plate_acc: Compute the number of license plates that were fully classified. For a single plate, if the ground truth is ABC123 and the prediction is also ABC123, it would score 1. However, if the prediction was ABD123, it would score 0, as not all characters were correctly classified.
cat_acc: Calculate the accuracy of individual characters within the license plates that were correctly classified. For example, if the correct label is ABC123 and the prediction is ABC133, it would yield a precision of 83.3% (5 out of 6 characters correctly classified), rather than 0% as in plate_acc, because it's not completely classified correctly.
top_3_k: Calculate how frequently the true character is included in the top-3 predictions (the three predictions with the highest probability).
plate_len_acc: Measures how often the predicted length of the license plate matches the ground truth. For example, if the target plate has 6 characters and the prediction also has 6, it scores 1 (regardless of content).

Region Metrics (When Enabled)¶

If your dataset includes a plate_region column and the plate config defines plate_regions, training enables a region head and logs:

region_acc: Categorical accuracy for region classification.
region_top3_acc: Top-3 categorical accuracy for region classification.
region_macro_f1: Macro-averaged F1 across all region classes.

Example Cases¶

Ground Truth	Prediction	plate_acc	char_acc	plate_len_acc	Notes
`ABC123`	`ABC123`	100%	100%	100%	Perfect match
`ABC123`	`ABD123`	0%	83.3%	100%	5 / 6 chars correct
`XYZ9`	`XYZ9`	100%	100%	100%	Short plate, all correct
`XYZ9`	`XYZ99`	0%	75.0%	0%	Length mismatch + one wrong
`ABC123`	`ABX1Y3`	0%	66.7%	100%	Two chars wrong