opencv 3.4.0 yolo implementation optimization [closed]

asked 2018-10-17 08:31:52 -0500

burakakde gravatar image

Dear all;

I am using yolo on a highly constrained system where GPU does not exist. So I have to live with CPU.

The first step that I have done was to implement darknet using pjreddie repository https://github.com/pjreddie/darknet

Then I did the same thing with AlexeyAB repo https://github.com/AlexeyAB

With the mentioned, the best FPS that I can get was 0.1.

Then I tried opencv implementation which is better optimized for CPU. With that I get 0.3 FPS.

The question is that is it possible to achieve higher FPS by using NNPACK, ARM_NEON (if using ARM) or any other optimization method in OPENCV implementation of darknet.

edit retag flag offensive reopen merge delete

Closed for the following reason question is off-topic or not relevant by berak
close date 2018-10-18 02:26:33.942572

Comments

i don't think, we can help you with your problem. this is the opencv QA.

(note, that this is NOT the opencv implementation of darknet (it's using opencv only for visualization))

((you CAN use pretrained yolo models with opencv's dnn module, but that was not, what your question was about, right ?))

berak gravatar imageberak ( 2018-10-17 08:47:43 -0500 )edit

@berak thx for the immediate reply. As you said that was not my question. I am already using the tiny model which is the fastest. I thought I can somehow optimize the yolo in opencv.

burakakde gravatar imageburakakde ( 2018-10-17 09:42:30 -0500 )edit

stop saying: " the yolo in opencv.". there is no such thing. (and we CAN'T help you with darknet)

berak gravatar imageberak ( 2018-10-17 09:52:20 -0500 )edit