Hi Raza!

A google search came up with some interesting references for you to start with:

https://www.sciencedirect.com/science/article/pii/S1574954123002431

https://www.sciencedirect.com/science/article/pii/S2351989422003523

Cheers,

Lars

I think re-id is quite different from a regular detection. I know that some great people in our Aalborg University have done this work on Giant Sunfish using Keypoint Mapping method. Perhaps, it could be useful for you to look at their methodology? :)



Link to ResearchGate: 

Direct link to pdf (university page): 

Also I've seen some sources doing re-id of spotted hyenas. I guess this would be more relevant for snow leopards as the pelt spot patterns can be used in both animal species for re-id. Paper here: 

P.S. Shameless plug here. We are considering to develop Re-id of certain species in our platform Animal Detect, as one of the features. :)

Hey Raza,

I would highly recommend for you to look into MegaDescriptor, which gives you the tools to quickly train your own model for re-ID (and its open source! 😊)

Paper:
https://openaccess.thecvf.com/content/WACV2024/papers/Cermak_WildlifeDatasets_An_Open-Source_Toolkit_for_Animal_Re-Identification_WACV_2024_paper.pdf

Github: https://github.com/WildlifeDatasets/wildlife-tools

HuggingFace: https://huggingface.co/BVRA/MegaDescriptor-T-224

If you wanna go and dig deeper, look into WildFusion: https://www.researchgate.net/publication/383413206_WildFusion_Individual_Animal_Identification_with_Calibrated_Similarity_Fusion which uses local and global matching to get better results than what’s shown on MegaDescriptor. The method @eugenegalaxy mentioned with sunfish uses a similar approach: https://www.mdpi.com/2077-1312/11/5/889 

If you are interested in individual re-ID, check out WildMe and their associated wildbooks! The HotSpotter model they use for African carnivores (like leopards) could probably be easily co-opted for snow leopards. 

Flagging @holmbergius who is the ED of WildMe!