How do keywords affect automated matching?

What Wildbook are you working in? ACW

How important is it to have the correct keywords on a media asset? If the keywords says “front” or “left” and the viewpoint is actually “right”, does this impact matches produced by the system to that annotation?

Just wondering how much attention we need to pay to the keywords the system generates and places on a media asset. It gets particularly complicated when there’s more than 1 animal in the media asset - how to make sure that the keywords are correct for each animal in the image?


TLDR; The keywords have no effect and are entirely ignored in the matching process.

Longer form:
The keywords are created by the detector for convenience but are completely duplicative of the viewpoint label on the annotation itself (e.g., "left, right, etc.). There is no filtering in matching based on keywords, so whether the keywords are present or absent on the MediaAsset is not consider.

Hope this helps.


1 Like