Compile error in module cnn_3dobj

[dbermond]

I'm getting the following compile error when trying to build OpenCV:

[62%] Building CXX object modules/cnn_3dobj/CMakeFiles/opencv_cnn_3dobj.dir/src/cnn_feature.cpp.o
/home/user/abs/opencv-noffmpeg-git/src/opencv_contrib/modules/cnn_3dobj/src/cnn_feature.cpp: In member function ‘void cv::cnn_3dobj::descriptorExtractor::loadNet(const cv::String&, const cv::String&, const cv::String&)’:
/home/user/abs/opencv-noffmpeg-git/src/opencv_contrib/modules/cnn_3dobj/src/cnn_feature.cpp:92:54: error: no matching function for call to ‘caffe::Net<float>::Net(const cv::String&, caffe::Phase)’
             convnet = new Net<float>(model_file, TEST);
                                                      ^

System information (version)

• OpenCV => git master (currently at 3.2.0.r332.gc1007c7276)
• OpenCV contrib modules => git master (currently at 3.2.0.r160.g58f7afcb)
• Caffe => rc5
• protobuf => 3.2.0
• OpenBLAS => 0.2.19
• Operating System / Platform => Arch Linux x86_64
• Compiler => gcc 6.3.1

Detailed description

The cnn_3dobj readme instructs to compile with -DWITH_QT:BOOL=OFF but this does not interfered in the error. Also, compiling with or without examples/tests did not make any difference regarding to this issue.
Caffe was manually compiled using OpenBLAS as the BLAS/CBLAS/Lapack provider and with CUDA support.

Steps to reproduce

Attached.
Complete output also provided.

opencv-cnn_3dobj-steps.txt
opencv-cnn_3dobj-output.txt

[alalek]

cnn_3dobj requires some exact version of Caffe and probably doesn't work with other versions.
Try to disable this module: cmake -DBUILD_opencv_cnn_3dobj=OFF ...

[dbermond]

@alalek
Oh, that's really sad to hear because I'm trying to build it with cnn_3dobj enabled.
I hope that a fix can be implemented so OpenCV can support Caffe latest versions including git master. Maybe the upcoming Caffe 1.0 release could be a good stimulus.

Anyway, thank you for the reply and support!

[sovrasov]

Probably the patch BVLC/caffe#5019 can resolve build problems, but it's still not accepted.