Create a folder for host the source

mkdir -p /opt/SP/packages/
cd $_

Clone the repository

git clone https://github.com/davisking/dlib.git
cd dlib
git submodule init
git submodule update

Create a folder for build the software (cmake have to be installed)

mkdir build
cd $_

Now run cmake with the following options:

cmake  -D DLIB_USE_CUDA=1 -D USE_AVX_INSTRUCTIONS=1 ../

If everything is fine, you can see a similar output

NOTE: Install cuDNN, openblas and Intel MKL/BLAS/LAPACK if your system is compliant with them.

Link to Intel Performance Library: https://software.seek.intel.com/performance-libraries

Link to cuDNN: https://developer.nvidia.com/cudnn

Now you can compile the source with

cmake --build . --config Release

After these step, the source is compiled.

Now you can install the python API.

cd ../
python setup.py install

NOTE: The compilation this time will use all the available CPU, be sure that you have enough memory for compile.

Once completed, you can query your python packages for be sure that dlib is now installed:

pip freeze | grep dlib

For check if the installation was succesfully:

Open a terminal and run the following

>>> import dlib.cuda as cuda;
>>> print(cuda.get_num_devices());
1
>>> import dlib
>>> dlib.DLIB_USE_BLAS
True
>>> dlib.DLIB_USE_CUDA
True
>>> dlib.DLIB_USE_LAPACK
True

Source: https://stackoverflow.com/a/57592670/9361998

@alessiosavi

After compiling with cmake --build . --config Release, I get this error (it's the same as before):

Building NVCC (Device) object dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cusolver_dlibapi.cu.o
nvcc fatal : Could not open output file /opt/SP/packages/dlib/build/dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cusolver_dlibapi.cu.o.NVCC-depend
CMake Error at dlib_generated_cusolver_dlibapi.cu.o.cmake:207 (message):
Error generating
/opt/SP/packages/dlib/build/dlib/CMakeFiles/dlib.dir/cuda/./dlib_generated_cusolver_dlibapi.cu.o

dlib/CMakeFiles/dlib.dir/build.make:70: recipe for target 'dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cusolver_dlibapi.cu.o' failed
make[2]: *** [dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cusolver_dlibapi.cu.o] Error 1
CMakeFiles/Makefile2:85: recipe for target 'dlib/CMakeFiles/dlib.dir/all' failed
make[1]: *** [dlib/CMakeFiles/dlib.dir/all] Error 2
Makefile:149: recipe for target 'all' failed
make: *** [all] Error 2

Have you run it as sudo?

@alessiosavi I experiencing the same issue as Azza71.
Using Ubuntu 18.04.4 bionic. CUDA 10.2. NVIDIA-SMI 440.33.01

I tried running

cmake --build . --config Release

and got this

[  1%] Building NVCC (Device) object dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cuda_dlib.cu.o
/home/alex/Desktop/dlib/dlib/cuda/cuda_dlib.cu(1691): 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.

1 error detected in the compilation of "/tmp/tmpxft_0000635c_00000000-6_cuda_dlib.cpp1.ii".
CMake Error at dlib_generated_cuda_dlib.cu.o.Release.cmake:279 (message):
  Error generating file
  /home/alex/Desktop/dlib/build/dlib/CMakeFiles/dlib.dir/cuda/./dlib_generated_cuda_dlib.cu.o


dlib/CMakeFiles/dlib.dir/build.make:583: recipe for target 'dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cuda_dlib.cu.o' failed
make[2]: *** [dlib/CMakeFiles/dlib.dir/cuda/dlib_generated_cuda_dlib.cu.o] Error 1
CMakeFiles/Makefile2:93: recipe for target 'dlib/CMakeFiles/dlib.dir/all' failed
make[1]: *** [dlib/CMakeFiles/dlib.dir/all] Error 2
Makefile:151: recipe for target 'all' failed
make: *** [all] Error 2

I tried running sudo cmake --build . --config Release , but it says cmake not found when sudo is used as well. cmake is installed on my computer.

try this command for cmake: cmake .. -DDLIB_USE_CUDA=1 -DUSE_AVX_INSTRUCTIONS=1 -DCUDA_NVCC_FLAGS="--expt-relaxed-constexpr"

Instead of cmake yourself, you can use this command in your root directory of dlib
python3 setup.py install --set DLIB_USE_CUDA=1 --set USE_AVX_INSTRUCTIONS=1 --set CUDA_NVCC_FLAGS="--expt-relaxed-constexpr"

This will automatically create the build folder.

Note: Whatever arguments you want to pass in cmake, you can directly pass in setup.py file as well
For example cmake .. -DDLIB_USE_CUDA=1 similar way in setup.py will be python3 setup.py install --set DLIB_USE_CUDA=1

Thank you all for the help. The issue was fixed when I looked at this, the issue exists for both jetson nano and jetson TX2: https://medium.com/@ageitgey/build-a-hardware-based-face-recognition-system-for-150-with-the-nvidia-jetson-nano-and-python-a25cb8c891fd