test one picture - Githubissues

SineXue commented 5 years ago

how to test one picture on this project, please

zhubenfu commented 5 years ago

1: 安装CUDA 9.2， cuda9.2版已经带有显卡驱动，默认安装，保持cuda和显卡驱动一致，否则会报cuda runtime is inefficient 35号错误。

2：确认各项目的include、lib和第三方库路径的配置，本人在项目中除CUDA使用绝对路径外，其余的include、lib路径均使用相对路径

caffe项目：
   C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\include

....\opensource\boost_1_57_0 ....\opensource\opencv\include ....\3rdparty\include ....\include ....\3rdparty\include\lmdb ....\3rdparty\include\hdf5 ....\src\ ....\3rdparty\include\glog ....\3rdparty\include\cudnn ....\src\caffe\proto ....\3rdparty\include\openblas

....\opensource\boost_1_57_0\lib ....\opensource\opencv\lib C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\lib\x64 ....\3rdparty\lib ....\tools_bin

libClassfication项目：
..\..\3rdparty\include\openblas

....\opensource\opencv\include ....\opensource\boost_1_57_0 ....\3rdparty\include ....\include ....\3rdparty\include\lmdb ....\3rdparty\include\hdf5 ....\src\ ....\3rdparty\include\glog ....\3rdparty\include\cudnn ....\src\caffe\proto C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\include

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\lib\x64 ....\opensource\boost_1_57_0\lib ....\opensource\opencv\lib ....\3rdparty\lib ....\tools_bin

ocr_test:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\include ....\opensource\boost_1_57_0 ....\opensource\opencv\include ..\libClassification ....\include ....\3rdparty\include ....\src\ ....\3rdparty\include\openblas

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\lib\x64 ....\opensource\opencv\lib ....\opensource\boost_1_57_0\lib ....\tools_bin 3：生成时要选择Release，x64格式

4：在项目的cuda生成host代码时，支持一些旧版本的cu伪代码，三个项目都设置成：compute_61,sm_61;compute_52,sm_52;compute_50,sm_50;compute_35,sm_35;compute_30,sm_30，否则会报no kernel image 40号错误

5：整理的3rdparty请使用群文件的3rdparty-20180730.tar

6：在visual studio 2015打开caffe.sln方案时（项目根目录下），如果出现无法加载工程的错误，打开项目根目录下的caffe-vsproj\caffe.sln

7：导入工程后，在ocr_test的项目，右键，选择【生成依赖项】-> 【项目依赖项】，勾选caffe, libClassfication两个项目

8: 项目代码中对路径没有做规范化处理，导致在运行期间报找不到模型文件的错误，为了调试方便，这里把代码中的路径全部修改成绝对路径，如ocr_test.cpp的962行， string modelfolder = "E:\License-Plate-Detect-Recognition\caffe-vsproj\ocr_test\plateCard_test"; 其他的类似修改但要区分MTCNN模型和ICNNPredict模型的位置

9：启动一个cmd，切换到工程根目录下的tools_bin目录，启动ocr_test.ext

Mebigger commented 5 years ago

@zhubenfu 您好，我在群里下的3rdparty20180726.7z（没有找到您说的3rdparty-20180730.tar）第三方文件， cuda改成了9.2，cudnn用7.1和3rdparty里边的都说不兼容！方便的话，您能告知一下用哪个版本的cudnn吗？十分感谢！

zhubenfu commented 5 years ago

链接：https://pan.baidu.com/s/1nX3CgpTckEOjSXlNvaruNw 提取码：8gls 或者加群群里有更新

zhubenfu / License-Plate-Detect-Recognition-via-Deep-Neural-Networks-accuracy-up-to-99.9

test one picture #13