Skip to content

ocr_torch是基于Torch1.8实现的DBNet(2.2M) + CRNN(3.8M)实现的轻量级文字检测识别项目(支持onnx推理).

Notifications You must be signed in to change notification settings

zeng-cy/ocr_torch

 
 

Repository files navigation

简介

ocr_torch是基于Torch1.8实现的DBNet(2.2M) + CRNN(3.8M)实现的轻量级文字检测识别项目(支持onnx推理).

项目环境
  • linux

  • python3.7

  • Torch1.8

文本检测模型DBNET

采用mobilenetV3 large作为骨干网络实现

  • 训练启动脚本
python -m torch.distributed.launch train.py -c config/train/det.yml
  • 测试启动脚本
python predict.py -c config/predict/det.yml
文本识别模型CRNN

采用mobilenetV3 small作为骨干网络实现

  • 训练启动脚本
python -m torch.distributed.launch train.py -c config/train/rec.yml
  • 测试测试脚本
python predict.py -c config/predict/rec.yml
文本检测识别合并推理
  • 训练推理脚本
python lite_ocr.py -c config/lite_ocr.yml
主要参考文献及源码
  1. DB https://github.com/MhLiao/DB
  2. PaddleOCR https://github.com/PaddlePaddle/PaddleOCR
  3. DBNET.pytorch https://github.com/WenmuZhou/DBNet.pytorch
  4. Paper https://arxiv.org/pdf/1911.08947.pdf

About

ocr_torch是基于Torch1.8实现的DBNet(2.2M) + CRNN(3.8M)实现的轻量级文字检测识别项目(支持onnx推理).

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%