search

jiexiong2016 / GCNv2_SLAM

Real-time SLAM system with deep features
Other
769 stars 194 forks source link

Trying to build for Raspberry Pi 4. Is this code CUDA only? #48

Open Fredrum opened 3 years ago

Fredrum commented 3 years ago

Hi, I was excited to find this as I'm trying to get a slam system to run 30fps on a Raspberry Pi 4! I just tried to build pytorch following your instructions, on my Raspian-32bit gcc8.3 raspberry.

If fails after ~10 minutes with the error messages,

Found a library with LAPACK API. (generic) disabling CUDA because NOT USE_CUDA is set -- CuDNN not found. Compiling without CuDNN support disabling ROCM because NOT USE_ROCM is set -- MIOpen not found. Compiling without MIOpen support

CMake Error at third_party/ideep/mkl-dnn/CMakeLists.txt:55 (message): Intel(R) MKL-DNN supports 64 bit platforms only

Does anyone know if I should be able to build and run this project?

Cheers Fred

cyizhuo commented 1 year ago

I built GCN in my Ubuntu-18.04 wsl with only cpu, it's slow (3fps max) but it did ran on cpu. Besides, use orb feature with GCN is 13fps max in my wsl. For your question, I did replace all kCUDA with kCPU in code, and cuda:0 with cpu in pt model files, but I never met your erroe.