if i’m understanding you correctly,
however, there are plenty of models that have been trained on iNaturalist data. so you might be able to play with those. for example, here are a few trained on iNat data from 2017: https://tfhub.dev/s?dataset=inaturalist&module-type=image-classification.
or else, you an train your own using the latest AWS Open Dataset: https://registry.opendata.aws/inaturalist-open-data/, or using the set from the 2021 Visipedia competition: https://github.com/visipedia/inat_comp/tree/master/2021