hua1024 / OpenOCR

Open OCR system, train and deploy

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

OpenOCR

简介

  • 通过自己搭建OCR的框架,总结学习到的OCR算法、pytorch的操作、python操作、部署操作
  • 同步添加部分通用目标检测模型供OCR场景使用
  • 性能正在逐步优化

更新日志

  • 添加amp和ema并测试通过 2021.04.17
  • TensorRt加速CRNN和DBNet测试通过 2021.03.22
  • DDP单机多卡测试通过 2021.01.18
  • 添加主干网络(ResNetVd、MobileNetV3) 2020.12.26
  • CRNN测试通过 2020.12.15
  • DBNet测试通过 2020.12.10
  • 框架训练测试推理测试通过 2020.11.30

目前已支持

  • DBNet
  • CRNN

复现记录

The ocr detect icdar2015 results should be as follows:

Model precision(ours) recall(ours) F-measure(ours) precision (paper) recall (paper) F-measure (paper)
DBNet-r50 88.9 77.6 82.9 88.3 77.9 82.8

TRT加速效果

https://github.com/hua1024/OpenDeploy

Reference

1.https://github.com/open-mmlab/mmdetection
2.https://github.com/PaddlePaddle/PaddleOCR
3.https://github.com/BADBADBADBOY/pytorchOCR

If this repository helps you,please star it. Thanks.

About

Open OCR system, train and deploy

License:Apache License 2.0


Languages

Language:C++ 52.4%Language:Python 47.3%Language:Cython 0.3%Language:Shell 0.0%Language:Makefile 0.0%Language:C 0.0%