A simple code for creating licence plate images and train e2e network
Branch: master
Clone or download
Latest commit 64490bc Feb 16, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
NoPlates new vision Feb 16, 2019
font new vision Feb 16, 2019
images new vision Feb 16, 2019
plate new vision Feb 16, 2019
plate_train new vision Feb 16, 2019
recognize_samples new vision Feb 16, 2019
.gitignore new vision Feb 16, 2019
PlateCommon.py new vision Feb 16, 2019
README.md new readme Feb 16, 2019
createImage_data.py new vision Feb 16, 2019
create_train_data.py new vision Feb 16, 2019
genplate.py new vision Feb 16, 2019
genplate_advanced.py new vision Feb 16, 2019
genplate_scene.py new vision Feb 16, 2019
plate_chars_set.txt new vision Feb 16, 2019
plate_size_china.md new vision Feb 16, 2019
test_nn.py new vision Feb 16, 2019
train_nn.py new vision Feb 16, 2019

README.md

hyperlpr-train_e2e

A simple code for creating licence plate images and train e2e network based on HyperLPR


Author LCorleone
E-mail lcorleone@foxmail.com

Requirements

  • tensorflow 1.5
  • keras 2.2.0
  • some common packages like numpy and so on.

Quick start

  • run create_train_data.py to create plate image and corresponding labels. This repository also contains the plate generator and can generate thousands of plates.
  • reset the train data path and run train_nn.py to train your model.

Attention

The image size created automatically is 120 * 30, fix the input size when you use the e2e network. You can create and train your own e2e network if you want.
Also, when tested in real scene, the e2e network performs not very well due to that the images' quality created automatically are still poor. If you have real image dataset and labels, it may be perfect.