This dataset contains two datasets. The first called plates has train, test and validation images for the 50 states. After running my dataset analysis and correction tool I found this dataset had duplicate images. I used that tool to create the new plates dataset that contains no duplicate images. I also added files for the District of Columbia(Washington DC) and 5 US territories so the new plates datset has a total of 56 classes. The new plates dataset is a very high quality dataset. All images are originals, no augmented images are present in the dataset. The dataset was carefully constructed so that it does not contain any duplicate images. . All images are size 128 X 224 X 3 in jpg format. All images have been cropped so the the license plate occupies at least 90% of the pixels in any image. This ensures that even fairly simple models will achieve high training, validation and test accuracy. Also included is a csv file so that users can use that to create their own train, validation and test sets if desired.