This model is released under the CC BY-NC-SA 4.0 licence (https://creativecommons.org/licenses/by-nc-sa/4.0/). It is based on EfficientNet V2 (Large).

This model is dedicated to the classification of the fauna from Kirghizistan. It was developped by Hex·Data (https://hex-data.io) on behalf of OSI-Panthera (https://www.osi-panthera.org/).

It is intended to be used on the crops produced by processing camera trap photos with MegaDetector (https://github.com/microsoft/CameraTraps/blob/main/megadetector.md). Humans and vehicles are expected to be detected by MegaDetector and thus are not handled by Panthera at all.

The model was trained on 42k images, with around 4k images per class, provided by the camera traps set up by OSI-Panthera.

The list of supported classes is available in the file "classes_Fri_Sep__1_18_50_55_2023.pickle". It contains a list of string, corresponding to the index used by the model for predictions. Here are the currently supported classes :

Class "vide" (empty, in French), is used to try to set aside the few false negatives returned by MegaDetector. The rest correspond to the scientific name of the family or species detected.

The following confusion matrix shows performances on the dev dataset. Columns with the "_incertain" prefix are meant for images where the confidence value was not high enough (<0.95).