Automation Rules

Help Animl help you

Automation Rules offer a flexible tool for configuring automated machine learning pipelines and email alerts. Users can configure what event they’d like the automation rule to fire on (either an image being added to the system or a specific label being added to an image), and then what action they’d like to happen in response (e.g., send an email alert, request ML inference from a specific ML model).

NOTE: Newly created Automation Rules will apply to all images added to your Project going forward. They do not apply retroactively to images that are already stored in your Project.

Creating an automation rule for ML Inference

To create an Automation Rule, elect the “Automation Rules” button (robot icon) from the left-hand sidebar, and fill out the form. A good initial rule to create is one that runs the Megadetector object detection algorithm on all images as they enter the system. To do so, complete form like so:

• Provide a name for the rule (e.g. “Request Megadetector prediction”)

• Trigger: “Image Added”

• Action: “Run inference”

• Model: “Megadetector”

• Confidence thresholds: 25% is a good default confidence for all for the classes predicted by Megadetector, but feel free to adjust if you find that Megadetector is missing a lot of a particular class (false negatives) or returning a lot of incorrect labels (false positives).

NOTE: You can also disable classes entirely by unchecking the checkbox to the left of them. This is useful in places in which you do not expect vehicles to be in any of the images, so having to manually remove false positives is time-consuming and undesirable.

Chaining together automation rules to create inference pipelines

Because one of the available “Trigger” events that automation rules can be set to execute on is “Label added”, and because when an ML model returns a prediction, it does in fact add a Label to a particular Image / Object, you can configure sequences of models to run your images through, depending on what labels were added in the previous inference step.

For example, if you have a fine-grained species classifier and you already have an automation rule set to request object detection from Megadetector on all new images (as described above), you can create a new rule to request predictions from the classifier only if Megadetector first returns an “animal” label, with the following settings:

• Name: anything really (e.g., “request classifier prediction”)

• Trigger: “Label Added”

• Label: “animal”

• Action: “Run inference”

• Model: “<name of classifier>”

• Confidence thresholds: Different models have different sensitivities, so getting confidence thresholds dialed in may require a bit of experimentation.

Creating an automation rule for Alerts

If you’d like to receive automated alert emails if a species of interest is detected, you can set an alert to send when a label for that species is added by an ML model:

• Name: anything really (e.g. “Rodent alert”)

• Trigger: “Label Added”

• Label: “rodent”

• Action: “Send Alert”

• To: “<user_a_email_address>, <user_b_email_address>”