WildTracker: Validating your species records and what we’re learning
WildTracker participants have now uploaded more than one million photos – an incredible collective effort that continues to grow. To better understand how accurate the classifications associated with some of those photos are, we recently kicked off a validation project in collaboration with the University of Tasmania.
The aim? To compare species tags assigned by landholders (that’s you!), those predicted by Stickybeak (our AI system), and those confirmed by third-party reviewers or experts. Understanding how closely these different sources agree (or don’t) helps us spot patterns, improve the system, and hopefully build better resources to help participants distinguish commonly misidentified species.

In May, we invited a group of TLC volunteers to help review a sample of tagged images. Their role is to confirm species ID, resolve conflicts between users and Stickybeak, and generally help clean the data. So, if you’ve noticed some past conflicts (!) between your tags and Stickybeak’s predictions quietly resolving into agreement (✓), that’s a direct result of this validation process.
One goal is to better understand where Stickybeak most often goes astray so that we can provide more targeted feedback to the smart people over at UTAS to improve Stickybeak’s accuracy and expand the list of species it can recognise. While it’s doing an impressive job on many common species – with agreement rates as high as 96% for wombats, native hens, pademelons, and wallabies – we know it has some quirks. The current model has shown an occasional enthusiasm for Lewin’s rail, fur seals, and Bassian thrushes where they don’t belong.

As a side benefit, the project is surfacing some excellent reference photos. We plan to use these in our Wildlife ID guide and at future WildTracker workshops. We’re also taking this opportunity to plan a tidy up “orphan tags”: species labels added before Stickybeak’s introduction in December last year that weren’t automatically paired to a bounding box on the animal in question.
We’ll be sharing more about this research at the TLC’s annual Conservation Science Symposium on 21 August. This event will be livestreamed; you can register here.