Image Embeddings Discussion

[celest atlas]

How to create meaningful image embeddings of fine grained objects such as fingerprints and natural phenomenon such as cyclones?
I have tried autoencoders results are average maybe need to evaluate dataset preparation.
Let's discuss on this. (Refer self supervised training methods Anuj Shah)

[young wing]

I don't know about fingerprints exactly, but for cyclones and natural phenomena you can refer to papers such as GraphCast, FourCastNet etc

As far as I remember, it's a combination of vision transformers, cross attention and some domain specific modules like fourier transformation or AFNO module