Hi,
I have compiled caffe (the version mentioned in this repo) successfully, and can also make matcaffe successfully, and build dcflow successfully following the instructions. However, I got the error below when I ran demo.m. I'm using matlab-R2016a version.
Invalid MEX-file '/home/project/dcflow/matlab/dcflow_mex.mexa64':
/home/project/dcflow/matlab/dcflow_mex.mexa64: undefined symbol: clGetProgramBuildInfo
Could you help? Thanks!
export INTELOCLSDKROOT or CUDA_PATH or AMDAPPSDKROOD is currently not working, as the compiler can't find the path of opencl (fatal error: CL/cl.hpp: No such file or directory)
A workaround is to set a global include path by export CPATH path/to/cl/include
Running the provided demo on both the KITTI and Sintel frames yields 0 matches; based on some preliminary investigation, it looks like the features computed by the provided convolutional net are identical for the two frames to compute the flow between (despite the frames themselves being different). This happens when running both the KITTI and Sintel frames.
I am not sure if this is due to some difference or error in my project setup -- are there specific versions of caffe or MATLAB that have to be used? Or for that matter, any specific versions of other dependencies (e.g. OpenCL)? My caffe and matcaffe build compiles and links without errors, and the dcflow code also builds and runs without errors, it just doesn't produce correct results.
TLDR are there any specific setup requirements, or any reasons for why the demo would not find any matches on the provided data?