#as for how many cameras with

1 messages · Page 1 of 1 (latest)

blissful pulsar
#

but no queues to handle a peaky load?

ionic needle
#

yes, queues are used

#

but for a single coral that is mostly moot

#

for multiple corals the queue becomes much more important

blissful pulsar
#

Stress testing: setting up a number cameras and subjecting them to a lot of motion and a lot and objects to see where and when when it breaks

ionic needle
#

yes, Blake has done that many times of the years he worked on frigate and again recently with 0.13 we did a lot of testing and iteration on object detection and motion detection making it more optimized and efficient

blissful pulsar
#

Cool.

#

I can't find a dropped frames sensor

#

looks like my max detection fps on one camera was 61

ionic needle
#

skipped fps

#

I actually have a graph on an HA dashboard that shows it for all cameras

blissful pulsar
#

Well a que would seen to eliminate dropped frames until the que is full even with one coral

ionic needle
#

no

#

because frigate is designed to run in real time

#

if it falls behind it will drop frames

#

because real time processing is the priority

blissful pulsar
#

at 6ms per inference for a coral if a second submission is made in less than 6 ms after a previous, it will be dropped or the process busy-waited.

ionic needle
#

ah no I don't think I was clear

#

I think we are describing the same thing in different terms

#

a queue is used, and it does have a max size

#

the max size is such that each camera does not fall behind more than 1 second from real time

blissful pulsar
#

ok that's reasonable. that's what I'd expect for an efficient process, is all the time critical code written in python or does it use compiled c/c++ libraries ?

ionic needle
#

it uses cython libraries like multiprocessing