-
-
Notifications
You must be signed in to change notification settings - Fork 91
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Processed events with no matches produce a lot of missed images #33
Comments
Maybe instead of having each detector pull the image, I write the image first and put that image through each detector. It could take a little longer, but would reduce the amount of images in general and Frigate API hits. |
I went ahead and made the changes I mentioned above in beta. Rather than each detector pulling images from Frigate on their own, the image is pulled then passed to each detector. This allows me to reduce the amount of images that are written. Hope this may help with the Frigate API issues too. This also sets up the code a little better to handle the MQTT snapshot issue you brought up - #18 |
cool! I actually wrote something similiar about only saving the best matches in #32 |
I'll switch over to beta again and help to try it out |
Awesome, thanks! One thing to note is the UI still displays the same image that was used per detector and if you delete one of the images that was used by another it'll delete both of them. I do want to clean this up still. Maybe displaying one "card" per image and then the results of each detector on that image. |
@erikarenhill let me know what you think about this. Since the same image is now passed to the detectors during each processing loop, I think it makes sense to display all the detection results on the card. It removes all the duplicate looking images when it was previously displayed by detector. Probably need to add a better way to determine which bounding box belongs to which result. |
@erikarenhill, curious to hear your thoughts on this as well.
I've been testing some of the recent code changes with untrained detectors. One thing I've noticed is that if
SAVE_UNKNOWN
is set totrue
, then a lot of images will be produced for each event. Each retry per detector could produce an image if a person is found. For example, if I'm running compreface and deepstack and the snapshot and latest retries are set to 10 each, then I could in theory have 40 images saved for just a single event.Maybe just taking a couple of the best results from the unknown images instead of saving out every single one to display on the UI.
The text was updated successfully, but these errors were encountered: