LPRNet Pytorch

Pytorch Implementation For LPRNet, A High Performance And Lightweight License Plate Recognition Framework.(Chinese Number Plates Recognition)

Indian Number Plate Recognition Modification.

Dependencies

• pytorch >= 1.0.0
• opencv-python 3.x
• python 3.x
• imutils
• Pillow
• numpy

Tasks

• Dataset preprocessor for csv label format.
• Added robustness to preprocessor.
• Tune hyperparameters.

Dataset preprocessing

1. Image name should be its label and separated into test and train. Otherwise:
2. Preprocessor.py will split data into train and test (9:1) and rename labels.
3. Run preprocessor.py and pass input folder, required format:
4. Input folder to contain 2 items, a folder containing all images and a csv/excel file of labels.
5. Csv should look like: | img name | Label | | :----: | :----: | | xyz.png | KA00XX0000 |

Training and Testing

1. Uncomment get_size function calls in train() to train with median size of dataset, default size is 94,24. Edit: Model only works for 94,24 size right now.
2. Based on your dataset path modify the script and its hyperparameters.
3. Adjust other hyperparameters if needed.
4. Run 'python train_LPRNet.py' or 'python test_LPRNet.py'.
5. If want to show testing result, add '--show true' or '--show 1' to run command.

References

1. LPRNet: License Plate Recognition via Deep Neural Networks
2. PyTorch中文文档
3. https://github.com/sirius-ai/LPRNet_Pytorch