It is fairly easy to check out and build dlib. Getting it to work in a performance-optimized manner – python bindings included -takes a little more work.<\/p>\n\n\n\n
Per the dlib github one can build the bindings by simply issuing:<\/p>\n\n\n\n
python setup.py install<\/code><\/pre>\n\n\n\nFirst problem I found is that the setup process decided to latch on to an old version of CUDA. That was my bad – fixed by moving my PATH variable to point to the new cuda’s bin dir.<\/p>\n\n\n\n
Second problem is that during compilation I saw the following:<\/p>\n\n\n\n
Invoking CMake build: 'cmake --build . --config Release -- -j12'\n[ 1%] Building NVCC (Device) object dlib_build\/CMakeFiles\/dlib.dir\/cuda\/dlib_generated_cusolver_dlibapi.cu.o\n[ 2%] Building NVCC (Device) object dlib_build\/CMakeFiles\/dlib.dir\/cuda\/dlib_generated_cuda_dlib.cu.o\n\/home\/carson\/code\/2020\/facenet\/dlib\/dlib\/cuda\/cuda_dlib.cu(1762): error: calling a constexpr __host__ function(\"log1p\") from a __device__ function(\"cuda_log1pexp\") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.\n<\/code><\/pre>\n\n\n\nAs it suggests this is resolved by passing in a flag to the compiler. To do this modify the setup.py line:<\/p>\n\n\n\n
python setup.py install --set USE_AVX_INSTRUCTIONS=1 --set DLIB_USE_CUDA=1 --set CUDA_NVCC_FLAGS=\"--expt-relaxed-constexpr\"\n<\/code><\/pre>\n\n\n\nEverything went just peachy from there except when I attempted to use dlib from within python I got an error (something like):<\/p>\n\n\n\n
dlib 19.19.99 is missing cblas_dtrsm symbol<\/code><\/pre>\n\n\n\nAfter which i tried importing face_recognition and got a segfault.<\/p>\n\n\n\n
I fixed this by install openblas-devel, then re-ran the setup.py script as above. Magically this fixed everything.<\/p>\n\n\n\n
Again, not bad – dlib seems cool – just normal troubleshooting stuff.<\/p>\n","protected":false},"excerpt":{"rendered":"
It is fairly easy to check out and build dlib. Getting it to work in a performance-optimized manner – python bindings included -takes a little more work. Per the dlib github one can build the bindings by simply issuing: First problem I found is that the setup process decided to latch on to an old … Continue reading How I Set Up DLIB<\/span>