make error in lib (on mac os 10.11.6 El Capitan) about tffrcnn

[pchankh]

Can we have an understanding what versions / requirements of each modules you are using for this build?

[eisneim]

i solved make error with these steps on my Mac Sierra

1. make sure cuda is in correct path, setup.py assumes cuda lib in '/usr/local/cuda/lib64', in my case, i put cuda lib in /usr/local/cuda/lib so i need to create a symbolic link use "ln -s /usr/local/cuda/lib /usr/local/cuda/lib64"

2. if you use python3 with tensorflow you need to install tensorflow for python2

3. if you encountered "error: no matching function for call to '_nms'"
   you need to manually change "nms/gpu_nms.cpp" file that's generated by cython.
   you need to cast long* to int* by adding "(int*)" to first argument like this:
   _nms((int*)(&(*_Pyx_BufPtrStrided1d( .......... (rest of code)........

4. if you encountered ld: symbol(s) not found for architecture x86_64 or nvcc fatal : Unknown option 'undefined'
   in lib/make.sh change nvcc compiler flags, remove $CXXFLAGS like this:

nvcc -std=c++11 -c -o roi_pooling_op.cu.o roi_pooling_op_gpu.cu.cc \
		-I $TF_INC -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC \
		-arch=sm_37

hope this will help

[CharlesShang]

@eisneim Thank you for your comments.