search

zhreshold / mxnet-ssd.cpp

C++ object detection module for mxnet-ssd
MIT License
97 stars 47 forks source link

how to enable GPU in the build #2

Open kaishijeng opened 7 years ago

kaishijeng commented 7 years ago

I don't see a flag to enable GPU in the install.sh. Do you know how to enable it?

Thanks,

zhreshold commented 7 years ago

This repo is intended for minimal dependency, enabling GPU is somehow against this goal. If you would like to use GPU, use python version instead please. If mxnet release building with cuda in amalgamation in the future, I will re-consider this problem again.

zhyj3038 commented 7 years ago

how fast mxnet-ssd.cpp detection object in image(eg.VOC2007) without GPU? thanks for your work very much!

zhreshold commented 7 years ago

Depending on the model/cpu/compilation mode, the run time significantly varies.

VGG16 model + 2.5GHz single thread with OPENMP disabled = 6s / image

yang-fei commented 6 years ago

I just wonder have you built this with GPU enabled? The demo instruction has the flag --gpu=INT. But when I run this, I got a error "D:\Dev\Projects\mxnet\dmlc-core\include\dmlc/logging.h:235: [19:14:50] D:\Dev\Projects\mxnet\src\storage\storage.cc:50: Please compile with CUDA enabled"

But I didn't see any place to enable compiling with CUDA enabled.

I have installed CUDA8.0 with cudnn v5.1 on Win10. NVIDIA Samples run well.

aa12356jm commented 6 years ago

@feiyang-nottingham the author alreally said: the repo don't support gpu