Vehicle Number plate detection using detectron2

This project demonstrates the use of Detectron2 by Facebook AI Research to detect and isolate number plates from vehicles.

Dataset: Custom dataset

Technology stacks and tools used

• Detectron2 by Facebook AI Research
• PyTorch deep learning framework.
• Cython
• Google Colab
• LabelMe
• CocoAPI

Installation

For PyTorch, use the command

pip install torch===1.4.0 torchvision===0.5.0 -f https://download.pytorch.org/whl/torch_stable.html

For Cython, use the command

pip3 install Cython

Then clone the detectron repo and cocoapi repo, using the commands

pip install -U 'git+https://github.com/facebookresearch/fvcore.git'
cd detectron2_repo
python setup.py build develop
git clone https://github.com/philferriere/cocoapi
cd .\cocoapi\PythonAPI\

Then unzip the plates_coc.zip in the main folder

Usage

First train the model using

python .\training.py

Then view the inferences by running

python .\inference.py   

There might be some packages missing in case you're not running this on colab. Mostly pertaining to missing packages, which can easily be identified

You can access a pretrained model here, in case you want to avoid training

Results

Training.py

Outputs of some images and their extracted plates

Considering that the dataset had a meager 414 images with annotations, the results were decent

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update the tests as appropriate.

References

• https://medium.com/towards-artificial-intelligence/number-plate-detection-recognition-detectron-v2-5ddad2a532d0

• https://github.com/facebookresearch/detectron2