/LPRNet_Pytorch

Pytorch Implementation For LPRNet, A High Performance And Lightweight License Plate Recognition Framework.

Primary LanguagePythonApache License 2.0Apache-2.0

LPRNet_Pytorch

Pytorch Implementation For LPRNet, A High Performance And Lightweight License Plate Recognition Framework.
完全适用于**车牌识别（Chinese License Plate Recognition）及国外车牌识别！
目前仅支持同时识别蓝牌和绿牌即新能源车牌等**车牌，但可通过扩展训练数据或微调支持其他类型车牌及提高识别准确率！

dependencies

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

pretrained model

training and testing

  1. prepare your datasets, image size must be 94x24.
  2. base on your datsets path modify the scripts its hyperparameters --train_img_dirs or --test_img_dirs.
  3. adjust other hyperparameters if need.
  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.

performance

  • personal test datasets.
  • include blue/green license plate.
  • images are very widely.
  • total test images number is 27320.
size personal test imgs(%) inference@gtx 1060(ms)
1.7M 96.0+ 0.5-

References

  1. LPRNet: License Plate Recognition via Deep Neural Networks
  2. PyTorch中文文档

部署依赖

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip3 install imutils

**车牌数据

CCPD

从训练数据中抽取文件作为测试数据

find . -type f -name 'Z*.jpg' | tail -n 1 | xargs -I {} mv {} /home/ye/CODE/MY/LPRNet_Pytorch/data/test/Brazil

常见问题