Hard to tell. Unfortunately neither this nor the openalpr project have any benchmarks or evaluations on a test set. It could be that the openalpr project alread

Hard to tell. Unfortunately neither this nor the openalpr project have any benchmarks or evaluations on a test set. It could be that the openalpr project already gives you 99% accuracy (unlikely), in which case there really isn't a point in trying to optimize it more (e.g. move on to harder problems).