How to skip frames in case of <30fps inference (MobileNet-SSD)

msharad19

Hi guys!
I'm running mobilenet-SSD for object detections on Oak-D Lite's camera feed. Since there seems to be some lag, I want to skip frames to enable real-time detection. How do I do this?
I already tried setting neural nets input to non-blocking:
net = pipeline.create(dai.node.MobileNetDetectionNetwork) net.input.setBlocking(False)
That doesn't seem to do the trick. I'm probably missing something in my understanding of pipeline creation then. Can someone please help me here?

Thanks!

erik

Hello msharad19 ,
The inference is about 100ms, so object detection results will always "lag behind" the original frame. Solution to this is frame/detection syncing, more info + demos can be found here. Thoughts?
Thanks, Erik

msharad19

Hello @erik
Thank you for this information. I will look into this. Thanks!