Skip to content

Detection

The failure detection feature is the most important feature of Forgetti.

Configuration

The configuration of the detection feature is fairly straightforward, and it is done almost entirely in the General widget of the configuration menu.

Camera URL

  • This is the URL of your camera stream.
  • Forgetti only supports MJPEG streams with programs like ustreamer or mjpeg-streamer.

Confidence Threshold

  • This is the minimum confidence required for Forgetti to respond to a failure.
  • For example, a confidence threshold of 60% forces Forgetti to only respond to failures with a confidence of 60% or higher.

CPU Threshold

  • This is the maximum CPU usage Forgetti will withstand before temporarily pausing inference.
  • This prevents Forgetti from crashing other processes such as Klipper.

Report Cooldown

  • This is the amount of time Forgetti will wait before reporting a second failure.
  • This is necessary to ensure that the same failure isn't reported more than once.

Model

  • This is the YOLOv8 model that Forgetti will use.
  • Although beefier models may have higher mAP scores, they have higher processing times and may not work on weaker hardware.
  • Bigger does not always mean better.
Model mAP Precision Recall
Nano 86% 92% 79%
Small 87% 91% 79%