msv2 vs msv3
What is the difference between conservationxlabs/miewid-msv2 and conservationxlabs/miewid-msv3? The model descriptions are identical excluding the results.
It would be great if some details about the training differences between the two models could be shared.
Thanks!
Seems like msv3 was trained on 64 datasets, whereas ms2 was trained on 54 datasets.
But I just read the READMEs:
MiewID-msv2 is a feature extractor trained for re-identification using contrastive learning on a large, high-quality dataset of 54 wildlife species - terrestrial and aquatic - including fins, flukes, flanks, faces.
MiewID-msv3 is a feature extractor trained for re-identification using contrastive learning on a large, high-quality dataset of 64 wildlife species - terrestrial and aquatic - including fins, flukes, flanks, faces.