Skip to content

Latest commit

 

History

History
123 lines (90 loc) · 7.84 KB

README.md

File metadata and controls

123 lines (90 loc) · 7.84 KB

«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

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

# 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/

Image1 Image 2 Image1 Image 2

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

Image1 Image 2 Image1 Image 2

Maintainers

  • zhujian - Initial work - zjykzj

Thanks

Contributing

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

Small note:

License

Apache License 2.0 © 2023 zjykzj