Help building cpp code with opencv

[ash_gamma]

I run into the following error when I try to compile a modified version of this project: GitHub - Wizaron/pytorch-cpp-inference: Serving PyTorch 1.0 Models as a Web Server in C++

I’m using the unzipped libtorch from Pytorch’s website. And here’s my CMakeLists.txt

make_minimum_required(VERSION 3.0 FATAL_ERROR)
project(cnn-classification)

find_package(Torch REQUIRED)
find_package(OpenCV REQUIRED)

add_executable(cnn-inference main.cc infer.cc ../../utils/opencvutils.cc ../../utils/torchutils.cc)

target_link_libraries(cnn-inference "${TORCH_LIBRARIES}")
target_link_libraries(cnn-inference "${OpenCV_LIBS}")

message(STATUS "OpenCV library status:")
message(STATUS "    version: ${OpenCV_VERSION}")
message(STATUS "    libraries: ${OpenCV_LIBS}")
message(STATUS "    include path: ${OpenCV_INCLUDE_DIRS}")


set_property(TARGET cnn-inference PROPERTY CXX_STANDARD 11)
#set_property(TARGET cnn-inference PROPERTY OUTPUT_NAME predict)

set(CMAKE_CXX_STANDARD_REQUIRED TRUE)

Could someone help me with these linkage errors? Do I need to compile libtorch from source?

Thanks!

[yf225 (PyTorch Developer, Meta)]

This can be due to ABI-incompatibility issue. You might want to try compiling libtorch from source, and make sure you are using the same compiler to compile both opencv and libtorch.

[yf225 (PyTorch Developer, Meta)]

We now have new ABI binaries for libtorch. They can be found on
http://pytorch.org, or at https://github.com/pytorch/pytorch/issues/17492#issuecomment-524692441. It should fix the problem.

[dambo (Shlomo)]

See my response here:

Libtorch with new C++11 ABI?

I have integrated the new ABI lib in my Docker file which also encapsulates the correct OpenCV version so that the notorious cv::imread() method works. I worked days and nights on attempting to compile libtorch + OpenCV from scratch with or without the -DCMAKE_CXX_FLAGS=-D_GLIBCXX_USE_CXX11_ABI=0 flag.

I used a pre-compiled OpenCV version. This is the only solution that worked for me.