Pytorch Implementation For LPRNet, A High Performance And Lightweight License Plate Recognition Framework.(Chinese Number Plates Recognition)
Indian Number Plate Recognition Modification.
- pytorch >= 1.0.0
- opencv-python 3.x
- python 3.x
- imutils
- Pillow
- numpy
- Dataset preprocessor for csv label format.
- Added robustness to preprocessor.
- Tune hyperparameters.
- Image name should be its label and separated into test and train. Otherwise:
- Preprocessor.py will split data into train and test (9:1) and rename labels.
- Run preprocessor.py and pass input folder, required format:
- Input folder to contain 2 items, a folder containing all images and a csv/excel file of labels.
- Csv should look like:
| img name | Label |
|---|---|
| xyz.png | KA00XX0000 |
- 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.
- Based on your dataset path modify the script and its hyperparameters.
- Adjust other hyperparameters if needed.
- Run 'python train_LPRNet.py' or 'python test_LPRNet.py'.
- If want to show testing result, add '--show true' or '--show 1' to run command.