alexvmt/TeraiNet

This model intends to support Bengal tiger conservation by classifying 10 categories of animals present in the Terai region of Nepal. The model was trained on about 2,000 images per class (source: LILA BC). The overall F1 score is 0.935 on a held-out test set (250 images per class), though performance varies somewhat by class. Local images from Nepal were not available so that images of visually similar 'substitute' species had to be used, i.e. the model's ability to generalize to the target region remains to be tested. The model is based on a pre-trained EfficientNetV2M. Image preparation (relying on MegaDetector) and model training follow MEWC.

The model can be used freely in AddaxAI.