Horama ID

Image classification is a mature technology useful for species identification by non-experts. At the same time, biodiversity collections are increasingly imaging their specimen collections, creating potential training image libraries for computer vision models.

A remaining bottleneck is the deployment of models to end-users such as other researchers, collection curators, or the public. Horama ID is a system intended to close the gap. It consists of a mobile app and a back-end that stores a collection of image classification models for species identification.

Taxonomists who want to contribute a model but are not AI experts themselves can train them on Custom Vision, which provides a graphic user interface and trains the model on Microsoft’s high-performance computing resources. The free tier allows the creation of two projects for up to fifty species. A detailed guide for contributors to Horama ID is available for download on this page.

End-users of identification models can install the Horama ID app on Android or iOS, download individual models, and switch between them as needed. Species profiles allow the verification of identification suggestions by the currently active model.

Horama ID is not intended to compete with broad applications like iNaturalist. Instead, it provides a means to deploy models of narrow taxonomic focus but comprehensive coverage, and potentially trained on preserved collection specimens, specific organs, or specific life stages such as herbarium specimens, pinned insects, or plant seeds.

Horama ID was funded by CSIRO and coded by 2pi Software of Bega, Australia.