Welcome to the License Plate Recognition demo project, powered by PaddleOCR!
This repository contains demo code for recognizing license plates, with detailed workflows and hands-on notebooks. The default environment targets Paperspace, but a Google Colab version is also available.
1. Start a Jupyter notebook in Paperspace.
2. Clone this repository:
git clone https://github.com/chunchet-ng/paddleocr_lpr.git3. Enter the project folder:
cd paddleocr_lpr4. Set up the Conda environment:
bash conda.sh5. Refresh your environment:
- Close all terminals and open a new one.
- Ensure the Conda
baseenvironment is activated.
6. Finalize setup:
bash setup.sh7. In Jupyter, select the paddleocr_lpr kernel for all notebooks.
-
Detection_Evaluation/HMean.ipynb
Calculate HMean for text detection and spotting tasks (with step-by-step examples). -
Recognition_Evaluation/Acc_Edit_Distance.ipynb
Compute accuracy and edit distance for text recognition tasks. -
License_Plate_Recognition/CCPD_2019.ipynb
Explore and preprocess the CCPD 2019 dataset for license plate detection and recognition. -
License_Plate_Recognition/EAST_CRNN_LPR.ipynb
Apply text detection (EAST), recognition (CRNN), and full license plate detection and recognition on CCPD 2019.
Note
Please run the CCPD_2019 notebook before starting with EAST_CRNN_LPR on your first use.
-
Text Detection & Spotting Evaluation:
How to calculate HMean with practical examples. -
Text Recognition Evaluation:
Methods for computing accuracy, edit distance, and counting correctly recognized words. -
Working with CCPD 2019:
- Convert the CCPD 2019 dataset into PaddleOCR format.
- Fine-tune and evaluate text detection (EAST) and recognition (CRNN) models.
- Compare pre-trained and fine-tuned models using the provided evaluation scripts.
-
CCPD 2019 dataset:
A large-scale, real-world Chinese city parking dataset for license plate detection and recognition. -
PaddleOCR:
Powerful and easy-to-use OCR tools.
- @nwjun: Special thanks for improving the notebooks and providing valuable feedback! 💪😇👍
Pull requests, bug reports, and suggestions are welcome! If you use this code or find it useful, please star the repo and share your results.