The problem
I want to optimize the processing time, I'm using darknet (melodic) Yolov4-tiny on a robot with CPU (i5-3610me) and 4 go ram before launching the node the 4 cores are running on ~70% (due to navigation and other stuff) and when I launch it, it goes up to 100% with 0.8 FPS
I can't run it on a GPU (CUDA is not supported on the robot)
I thought about downgrading the Yolo version but I will lose in the detection
I thought about launching the darknet on only some frames meaning the darknet node "sleeps" for some time but it is not helpful for the streaming case
Idea
I thought about processing some portion of the frame meaning the darknet input (the topic of /rgb/image_raw) will take half of the frame maybe that could help improve a little bit, if so how can I try it?
I want this issue to be an active issue for the community to share their ideas on how we can optimize for this kind of situation
The problem I want to optimize the processing time, I'm using darknet (melodic) Yolov4-tiny on a robot with CPU (i5-3610me) and 4 go ram before launching the node the 4 cores are running on ~70% (due to navigation and other stuff) and when I launch it, it goes up to 100% with 0.8 FPS
Idea I thought about processing some portion of the frame meaning the darknet input (the topic of /rgb/image_raw) will take half of the frame maybe that could help improve a little bit, if so how can I try it? I want this issue to be an active issue for the community to share their ideas on how we can optimize for this kind of situation
Thank you