0
$\begingroup$

Is it possible to run SSD or YOLO object detection on raspberry pi 3 for live object detection (2/4frames x second)? I've tried this SSD implementation but it takes 14 s per frame. Is there anything I could do to speed up ?

$\endgroup$
3
$\begingroup$

No. YOLO and SSD are based on Nvidia's proprietary CUDA technology which is not available on Raspberry simply because of the GPU vendor is not Nvidia. Even more, there seems to be no implementation of even OpenCL for the Raspberry's GPU. What you can do is to try port YOLO's of SSD's CNN core from CUDA to Raspberry GPU's assembler, in the way described in, for example, https://rpiplayground.wordpress.com/2014/05/03/hacking-the-gpu-for-fun-and-profit-pt-1/

$\endgroup$
0
$\begingroup$

See this github repo: https://github.com/01org/caffe This is an OpenCL implementation of Caffe. If you take a look at the Wiki page for the "Inference optimization" branch at https://github.com/01org/caffe/wiki/clCaffe you will notice that you can convert a YOLOv2 model to a Caffe model and run it under OpenCL. Porting CaffeCL to ARM should be straightforward as it already works on several OpenCL platforms.

$\endgroup$
0
$\begingroup$

Try nnpack, I can run tiny yolo @ approx 1 sec /frame on ARM based CPU.

https://github.com/digitalbrain79/darknet-nnpack

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.