Can you explain the bufferring for synchronous same frame?

jaiminLee

I don't know how to fix it.

I did gather color frames and I processed face estimation.

However, it gradually takes longer to find the face position.

The number of (color frame and face detect frame) getSequenceNum() is the same.

You can see attached my picture. When I shake my head I can see that phenomenon.

Facial recognition follows behind. It has to process the recent frame.

I need your advice.

erik

Hi jaiminLee ,
I think it's a problem as the face detection (NN) results / tracking results aren't being synced with the color frame they are being inferenced upon - so they lag behind. Docs on frame syncing here. For it to "gradually take longer", I assume it's due to queue sizes, and them filling up, so perhaps decreasing those and setting to non-blocking (docs here) would be better. Thoughts?
Thanks, Erik

jaiminLee

erik
Ok. It is good.
I did like this.
auto preview = device.getOutputQueue("preview", 1, false);
auto tracklets = device.getOutputQueue("tracklets", 1, false);
auto xoutBoundingBoxDepthMappingQueue = device.getOutputQueue("boundingBoxDepthMapping", 1, false);
auto depthQueue = device.getOutputQueue("depth", 1, false);
auto faceResultOutQueue = device.getOutputQueue("faceResultOut", 1, false);
auto faceRecOutQueue = device.getOutputQueue("recognition", 1, false);
So. I can't see any more delay.
Thank you so much.

erik

Hi jaiminLee ,
Besides host side queues, you also have "device side / node queues", which can also cause the latency issues. THanks for confirming it's better now!
Kind regards, Erik