README.md

«LPDet» provides a complete License Plate Detection and Recognition algorithm

ONLINE DEMO：LICENSE PLATE DETECT/SEGMENT/RECOG

Implementing license plate detection, segmentation, and recognition functions based on ultralytics/yolov5 v7.0 and zjykzj/crnn-ctc

Model Segmentation	Input Shape	GFLOPs	Model Size (MB)	Speed RTX 3090 b1 (ms)	ChineseLicensePlate mAP50 (%)	Training Data	Testing Data
YOLOv5n-Seg	(3, 640, 640)	6.7	3.9	9.0	99.2	200,579	105,585
Model Recognition	Input Shape	GFLOPs	Model Size (MB)	Speed RTX 3090 b1 (ms)	ChineseLicensePlate Accuracy (%)	Training Data	Testing Data
CRNN_Tiny	(3, 48, 168)	0.3	4.0	7.5	76.226	269,621	149,002

Table of Contents

• Table of Contents
• News🚀🚀🚀
• Background
• Usage
  • Detect + Seg
  • Recog
• Maintainers
• Thanks
• Contributing
• License

News🚀🚀🚀

Version	Release Date	Major Updates
v1.2.0	2024/08/17	Add ONNX inference and Gradio demo.
v1.1.0	2024/08/04	Optimize license plate segmentation and recognition algorithms.
v1.0.0	2024/07/21	Implementing license plate detection, segmentation, and recognition functions.
v0.3.0	2023/10/03	Support for Automatic Mixed Precision (AMP) training.
v0.2.0	2023/10/02	Support for distributed training with multi-GPUs.
v0.1.0	2023/09/29	Reconstruct the 872699467/CCPD_CNN implementation to adapt to interfaces after Pytorch v1.0.0.

Background

This warehouse provides a complete license plate detection and recognition algorithm, with the goal of perfectly detecting and recognizing all license plates and license plate information.

Note: the latest implementation in our warehouse is entirely based on ultralytics/yolov5 v7.0

Usage

Detect + Seg

• ChineseLicensePlate: Baidu Drive(ad7l)

# Train
$ python segment/train.py --data ChineseLicensePlate-seg.yaml --weights yolov5n-seg.pt --img 640 --epoch 10
# Eval
$ python segment/val.py --weights yolov5n-seg_plate.pt --data ChineseLicensePlate-seg.yaml --img 640
# Predict
$ python segment/predict.py --weights yolov5n-seg_plate.pt --source ./assets/ccpd/

Recog

About license plate recognition algorithm, using zjykzj/crnn-ctc

$ git submodule init
$ git submodule update

Then predicting license plates

# Using Pytorch
$ python3 segment/predict_plate.py --weights yolov5n-seg_plate.pt --w-for-recog crnn_tiny-plate-b512-e100.pth --source ./assets/ccpd/
# Using ONNXRuntime
$ python3 segment/predict_plate.py --weights yolov5n-seg_plate.onnx --w-for-recog crnn_tiny-plate.onnx --source ./assets/ccpd/ --device cpu

Maintainers

• zhujian - Initial work - zjykzj

Thanks

• detectRecog/CCPD
• 872699467/CCPD_CNN
• zjykzj/FastestDet
• zjykzj/YOLOv5
• zjykzj/crnn-ctc
• ultralytics/yolov5

Contributing

Anyone's participation is welcome! Open an issue or submit PRs.

Small note:

• Git submission specifications should be complied with Conventional Commits
• If versioned, please conform to the Semantic Versioning 2.0.0 specification
• If editing the README, please conform to the standard-readme specification.

License

Apache License 2.0 © 2023 zjykzj