Deepstream multi FPS inference with no sync

Please provide complete information as applicable to your setup.

• Hardware Platform Jetson AGX XAVIER
• DeepStream Version 6.2
• JetPack Version 5.1

Hi I want to infer on multiple input streams that have dissimilar FPS (e.g. 30 and 9) , I do not want batch synchronisation enforced and I want to let inference flow freely with old frames latched (in case the batch cannot be filled due to some frames arriving late)

basically I want to process frames as they come in, in the end I want to use sticky queues (where I can latch old frames to fill the intervals where the slow source can;t keep up with the fast source) and a tiler to stich the frames and display as a grid.

currently it seems the output is bottlenecked to the rate of the slowest FPS source, I think this may be due to efforts of holding back quicker inferences buffers until slow frames arrive.

Even if I demux (split) sources and infer with batch size one inference this slowness seems to not go away.

Can you suggest somehting for this please, this is just a demo and not a mission critical scenario (where we do not need the addiitonal overheads that bogs down the pipeline) so even a bit jankey solution would help.


P.S currently trying to use the videorate element

It appears to be that videorate was the hero we needed after all! :)

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.