Trained on a dataset of Bluey 512x512 images. This model works best when paired alongside it's embedding(s)!
I had planned to refine this further but didn't get around to it, expect a variety of scuffed results.