Page 2 of 2

Re: Need testers for ES code (multiple model chaining/ arbitrary frames)

Posted: Wed Jan 06, 2021 8:28 am
by Magic919
I've pulled the latest version and moved to that and the migated config now.

Re: Need testers for ES code (multiple model chaining/ arbitrary frames)

Posted: Wed Jan 06, 2021 11:38 am
by Magic919
Contradition regarding the default for use_sequence. The config file says

Code: Select all

 NEW: If enabled (default = no), we will use ml_sequence and stream_sequence
Readthedocs says

Code: Select all

When use_sequence is set to yes (default)

Re: Need testers for ES code (multiple model chaining/ arbitrary frames)

Posted: Wed Jan 06, 2021 3:57 pm
by asker
Magic919 wrote:
Wed Jan 06, 2021 11:38 am
Contradition regarding the default for use_sequence. The config file says

Code: Select all

 NEW: If enabled (default = no), we will use ml_sequence and stream_sequence
Readthedocs says

Code: Select all

When use_sequence is set to yes (default)
fixed.

Re: Need testers for ES code (multiple model chaining/ arbitrary frames)

Posted: Wed Jan 06, 2021 4:15 pm
by Magic919
Thanks again. ZMES is exceeding my needs, but it's great that you've carried on extending it.

Re: Need testers for ES code (multiple model chaining/ arbitrary frames)

Posted: Wed Jan 06, 2021 5:05 pm
by ranger21
asker wrote:
Wed Dec 30, 2020 9:33 pm
The new version of ES support model chaining as well as inspecting multiple frames in an event.
I am looking for folks to:

1. Update to the new ES
2. Test if current operation works without making any config changes (that is, new code with old functionality)
3. Test new functionality.

Specifically, I want to know if folks update to the new ES and face any errors (without making changes to config)

Steps are here: https://github.com/pliablepixels/zmeven ... -749162186

Would appreciate feedback
default config for new detection sequence makes it to analyse faces all the time, so my cpu utilization sky-rocketed, i've added to face model config of ml_sequence "pre_existing_labels": person and now it's back to normal

it's not possible to setup different ml_sequences for certain monitors?

well, i have only CPU, so i can't test coral edge or gpu, but maybe i will try to do some ML_SEQUENCE with tiny_yolov4 and higher confidence number, then full yolo4 with lower confidence, but im not really memory bound.

Re: Need testers for ES code (multiple model chaining/ arbitrary frames)

Posted: Wed Jan 06, 2021 5:53 pm
by asker
default config for new detection sequence makes it to analyse faces all the time, so my cpu utilization sky-rocketed, i've added to face model config of ml_sequence "pre_existing_labels": person and now it's back to normal
The default config is a sample. If you don't want face, simply remove it from model_sequence
it's not possible to setup different ml_sequences for certain monitors?
It should work. You should be able to replace either variables or the full ml_sequence on a per monitor basis. Have you tried?

Re: Need testers for ES code (multiple model chaining/ arbitrary frames)

Posted: Sat Jan 09, 2021 4:24 pm
by ranger21
Yeah, global variables from [custom] section + replacing in monitors worked fine. Thank you.