Curricular LPR - A transformation of an existing synthetic LPR dataset for curricular learning, inspired by the article "CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition" (https://arxiv.org/abs/2004.00288). Images contain various levels of blur from an original dataset using Pillow.

Purpose: Training a LPR framework in curricular learning: Starting with some number of epochs on regular images only, then adding the first level of blur for X epochs, and then adding the second level of blur for Y epochs.

Data: There are 19180 images in total, after removing some badly generated images from the original synthetic dataset. These are divided into three levels of blur. 10000 - No blur at all 6000 - Blur level 3 (or 1 for small images) 3180 - Blur level 5 (or 2 for small images) Blur was chosen randomly for each image between BoxBlur and GaussianBlur

Source: https://www.kaggle.com/datasets/nickyazdani/license-plate-text-recognition-dataset