A PyTorch implementation of End to End Chinese License Plate Recognition, the whole project is mainly modified from SSD, which is a Single-stage object detector. And I mainly adopted the idea of this article(A Novel Integrated Framework for Learning both Text Detection and Recognition) and designed an OCR ROI pooling operation, so that detection, classification and text recognition can be integrated into one net for training, and using VGG as the backbone network.
And i use CCPD: Chinese City Parking Dataset for training, i random choose 70000 sample for training. You can download the whole dataset or just use the images in ./data/test_data
- clone the project
git clone https://github.com/chenjun2hao/CLPR.pytorch.git- install cupy, for example:
(Binary Package for CUDA 9.0)
$ pip install cupy-cuda90- Other dependencies
python 3.6
pytorch 0.4.0For testing, download the pretrained model from Baidu disk, password:16pk, and put the model in weights folder.
python test.py
here are some examples:
 图1 |
 图2 |
 图3 |
 图4 |
 图5 |
 图6 |
Coming later
- [] For strabismus license plate recognition effect is not good, can add amendments
- [] Character recognition is easy to leak out, and it combines multi-layer features for character recognition.