asked 2017-12-11 03:52:36 -0500

wingsoflight gravatar image

Hello, I recently tested pretrained MobileNet SSD on my machine by using cv2.dnn module in python. It seems my code is only computing on CPU. So, I wanted to know: is there is any GPU support in cv2.dnn module?

edit retag flag offensive close merge delete


Not yet ... its planned for the near future, but for now, no GPU support yet.

StevenPuttemans gravatar imageStevenPuttemans ( 2017-12-11 06:04:33 -0500 )edit

I'm really happy about the DNN module - especially because the DNN libraries have very limited C++ support. However it's really slow on CPU, so I also hope that it will be implemented on GPUs too - so you get my upvote.

OTOH as long as you use Python, you can use the original DNN library for processing - they all have good GPU support.

kbarni gravatar imagekbarni ( 2017-12-11 06:36:58 -0500 )edit
dkurt gravatar imagedkurt ( 2017-12-11 07:28:39 -0500 )edit

Great! (I prefer OpenCL support, as it's much more general than Cuda)

kbarni gravatar imagekbarni ( 2017-12-11 08:18:31 -0500 )edit

I was able to run my script on my GPU by using OpenCL, but it runs 2x slower than on CPU :\

wingsoflight gravatar imagewingsoflight ( 2017-12-12 02:41:37 -0500 )edit

@wingsoflight, what kind of GPU?

dkurt gravatar imagedkurt ( 2017-12-12 08:53:27 -0500 )edit

@dkurt, Quadro M2000

wingsoflight gravatar imagewingsoflight ( 2017-12-12 14:39:03 -0500 )edit

Where you able to figure out a solution for the poor performance on GPU?

ashr gravatar imageashr ( 2018-05-09 01:51:24 -0500 )edit

Did you figure out why it ran slower on the GPU when compared to the CPU? Isn't it supposed to be the other way around?

tejasa97 gravatar imagetejasa97 ( 2018-08-18 03:09:29 -0500 )edit