Skip to content

A pytorch re-implementation of Real-time Scene Text Detection with Differentiable Binarization,and use the openvino inference engine for inference in the prediction phase

License

Notifications You must be signed in to change notification settings

DYF-AI/DBNet-OpenVINO

 
 

Repository files navigation

Real-time Scene Text Detection with Differentiable Binarization

note: 原始版本 DBNet.pytorch

中文解读

network

安装环境

请参考原始版本的Readme

How to run openvino

1.下载代码训练的模型地址:渣云:访问密码 myj4 ,将模型放到onnx-xml文件夹 (仅供演示,该模型未训练完全) 2.执行 predict_openvino.sh

  • 不包含预处理和后处理时间(不同的CPU推理时间会存在差异)
time(ms) pytorch onnxrt openvino
inference 400ms 278ms 150ms

修改之处

本repo为了可以使用tensorRT加速,将反卷积操作全部改为upsample。比如

 # 原始版本
nn.ConvTranspose2d(in_channels // 4, in_channels // 4, 2, 2), # 上采样两倍
# 修改版本 
nn.Upsample(scale_factor=2, mode='nearest'),

更多的修改,请看代码:

models/head/DBHead.py 
models/model.py 
models/neck/FPN.py

模型

修改后代码训练的模型地址:渣云:访问密码 myj4

目前没有训练完成,相比原版模型(1200epoch),只训练了500epoch。精度:90.0 召回率:68.2。

可以自己去训练。

TensorRT版本

https://github.com/BaofengZan/DBNet-TensorRT

About

A pytorch re-implementation of Real-time Scene Text Detection with Differentiable Binarization,and use the openvino inference engine for inference in the prediction phase

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.5%
  • Shell 0.5%