Awesome Open Source
Awesome Open Source

end-to-end-for-plate-recognition

多标签分类,端到端的中文车牌识别基于mxnet . 从xlvector的ocr代码修改,减少了参数,由于我没有显卡。单线程 9 samples/s 速度 ,用CPU在MBP上跑了50w张样本。识别率到了81%。不过还没有完全收敛。

训练好的模型

https://awesomeopensource.com/project/ibyte2011/end-to-end-for-chinese-plate-recognition

关于车牌识别

生成的车牌对于实际车牌并不是效果很好,在结合真实样本和GAN,训练了一个更好的模型,对真实车牌表现很好。 并实现了一整套车牌识别的系统命名为HyperLPR https://awesomeopensource.com/project/zeusees/HyperLPR

依赖:

  • Numpy
  • Mxnet
  • Opencv

生成的车牌样张

通过渲染车牌加上畸变、噪声、与自然环境结合生成车牌的样本。

image image image image image image image image image image image image

识别样张

Author

  • Jack Yu
  • Xiao Xiao
Alternatives To End To End For Chinese Plate Recognition
Select To Compare


Alternative Project Comparisons
Related Awesome Lists
Top Programming Languages

Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
Python (888,878
Opencv (13,681
Recognition (10,744
Chinese (6,477
Numpy (6,275
Generative Adversarial Network (5,357
Mxnet (1,232