Automatic License Plate Generation and Detection NOTE: the evaluation side of the pipeline is still under development

LPGenerator.py - Python Script to generate image sets of randomized Maryland License Plate Sequences

MatlabImplementation - All Matlab Development Code

FromScratch - Python (Keras/Tensorflow) Development and Testing

Archived - Old files to keep around, not under active development or use other than as refrence

1. Clone this repository git clone https://github.com/Hriste/AutomaticLP.git (if on MARCC use the /code folder)
2. In the top level of this repository clone the tensorflow object detection repository git clone https://github.com/tensorflow/models
3. Modify the ~./bashrc file with the following 2 lines where pwd is the path to the tensorflow object detection repository models/research folder.

export PYTHONPATH=$PYTHONPATH:`pwd`
export PYTHONPATH=$PYTHONPATH:`pwd`/slim

4. From the top level of the AutomaticLP Repository run the setUpScript.sh
   If you have issues with this run the commands individually.
   NOTE: the script assummes you're executing from the user@jhu.edu/code/AutomaticLP if this is not the case adjust the change directory commands in the script accordingly.

Video Training Walkthrough HERE

7. Run jupyter lab on a partition with a gpu - I usually use the following

sbatch -p debug -c 6 --gres=gpu:1 -t 2:0:0 jupyter_marcc lab

8. Launch the GPUTraining notebook

   • Make sure to select the LPKernel.

9. Running the notebook will populate the trainingOutput directory - this directory can then be used with tensorboard to evaluate results.
   If you want a brand new training session delete any prexisting trainingOutput folder.
   Each time you run reset the Kernel - otherwise there can be issues logging to file

[ ] Add Evaluation to pipeline [ ] Add setup needed for evaluation to run (import thrid party projects, add packages, etc.)

We need a version 1 installation of tensorflow NOT 2.*.

If you have tensorflow already installed check the version by running python in a terminal window and running:

import tensorflow as tf
print(tf.__version__)

If your version of tensorflow is 2.* run 'pip uninstall tensorflow'. If you don't have tensorflow installed or have just unistalled a 2.* version run 'pip install tensorflow=1.*'

(I am working with tensorflow 1.15.2)

On MARCC we're using a conda enviroment so instead of using pip uninstall / install use conda install

Using the tensorflow object detection API training and evaluation is driven by a configuration file.

The configuration file is located in models/model/ssd_mobilenet_v1_coco.config

(note very little training was done for this but I wanted to put a sample here)

Training Procedure based on THIS

Legacy Training (train.py) based on THIS and THIS

Exporting Models

Installation Instructions

[Tutorial 1] (https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/training.html#monitor-training-job-progress-using-tensorboard)

Tutorial 2