Single-row License Plate Recognition using Convolutional-Recurrent Neural Network (CRNN)

Usage of Pre-trained License Plate Recognition Model

git clone https://github.com/kfengtee/crnn-license-plate-OCR.git
cd crnn-license-plate-OCR
pip install -r requirements.txt

import model.alpr as alpr

# create ALPR instance (change parameters according to needs)
lpr = alpr.AutoLPR(decoder='bestPath', normalise=True)

# load model (change parameters according to needs)
lpr.load(crnn_path='model/weights/best-fyp-improved.pth')

# inferencing
lpr.predict('path/to/image')

Test Model Performance on New Dataset

python test.py --crnnPath path/to/pretrained/weights --dataPath path/to/test/data --savePath path/to/save/results

Optional arguments:

1. --ctcDecoder : ['bestPath' or 'beamSearch'],
   • Method to decode CTC output.
   • Source: https://towardsdatascience.com/beam-search-decoding-in-ctc-trained-neural-networks-5a889a3d85a7
2. --normalise : boolean,
   • Whether to normalise the posterior probability with prior probability or not (to avoid bias).

Incremental Training / Retrain Model with Own Dataset

python train.py --dataPath path/to/training/data --savePath path/to/save/model 

To know more about the tunable hyperparameters

python ./train.py --help