FPS values at multiple detections

martin181818

I am working on depthai-experiments/gen2-age-gender/api at master · luxonis/depthai-experiments (github.com).
I am using my own detection model and it is working fine for one or two detected objects.
The problem I am facing is when the number of detection per image surpass 4 or 5 objects: the fps value decreases below 4. I just want to understand in this case what is memory consuming in this case and are there any possibility to increase the FPS when detecting multiple objects in the same frame ?

Thank you very much

erik

Hi martin181818 ,
For each face on the image you are doing one additional infernece (besides the face-detection one). So with 5 faces on frame, you are doing 1 (detection) + 5 (recognition) inferences for each frame. Perhaps try the sdk implementation - I believe it should skip inferencing for some frames if it can't keep up (too many faces). Thoughts?
Thanks, Erik

martin181818

Thank you for your reply. I will check it