Do We Train on Test Data? The Impact of Near-Duplicates on License Plate Recognition

TL;DR

This paper reveals that near-duplicate images in license plate recognition datasets inflate model performance metrics, and demonstrates that removing these duplicates leads to more accurate evaluation and model ranking.

Contribution

It identifies the prevalence of near-duplicates in LPR datasets and provides a methodology for creating fair, duplicate-free splits to improve evaluation integrity.

Findings

Recognition rates drop significantly without duplicates.

Model rankings change when evaluated on fair splits.

Near-duplicates bias the assessment of LPR models.

Abstract

This work draws attention to the large fraction of near-duplicates in the training and test sets of datasets widely adopted in License Plate Recognition (LPR) research. These duplicates refer to images that, although different, show the same license plate. Our experiments, conducted on the two most popular datasets in the field, show a substantial decrease in recognition rate when six well-known models are trained and tested under fair splits, that is, in the absence of duplicates in the training and test sets. Moreover, in one of the datasets, the ranking of models changed considerably when they were trained and tested under duplicate-free splits. These findings suggest that such duplicates have significantly biased the evaluation and development of deep learning-based models for LPR. The list of near-duplicates we have found and proposals for fair splits are publicly available for further research at https://raysonlaroca.github.io/supp/lpr-train-on-test/