This repository helps you generate a veritable cornucopia of persian licence plates. You're almost done with this part, training your own CNN would be the next step though!

Generated sample:

For the sake of ease, generated licence plates come with their annotations:

Generated sample with perspective transformations:

On top of all that, it automatically generates associated xml files in pascal voc format:

Sample neural network trained on this dataset:

Model zoo faster_rcnn_inception_v2_coco

Because of oop pattern you won't need to make significant changes unless you want something structurally different!

*Note: if you have trouble installing jsonnet on windows 
use win_conf.json instead and remove jsonnet dependencies 

1. Got all requirements installed:
   pip3 install --upgrade -r requirements.txt

2. Tweak configuration file: (optional)

   • project_configurations.jsonnet

3. Out-of-the-box data generation:

   • With default parameters:
     $ python3 main.py
   • Override config parameters:
     $ pyhton3 main.py --num_out_img 10000
$ pyhton3 main.py --apply_misc_noise False
$ pyhton3 main.py --apply_dirt False
$ pyhton3 main.py --output_directory 'output' (doesn't already exist)
     $ pyhton3 main.py --img_per_package 8000