[BayramHm]

Hello my friends, i'm working with SuperGlue is a CVPR 2020 research project done at Magic Leap. The SuperGlue network is a Graph Neural Network combined with an Optimal Matching layer that is trained to perform matching on two sets of sparse image features. This repo includes PyTorch code and pretrained weights for running the SuperGlue matching network on top of [SuperPoint](https://arxiv.org/abs/1712.07629) keypoints and descriptors. i'm wondering if it is possible to pass from SIFT to DEEP LEARNING when i compare a pair of image to show difference between them ?
Or if anyone can help me with an AI algorithm that compare a pair of image to show difference between them.
Thanks for your sharing.

[frankinstien]

To allow for that kind of comparison CNNs have a problem in that they look for labeled classifications. What is needed is a means to generalized visual features at a basic level of complexity or information content. The generalizations can allow for ambiguity in terms of exact details of bits of visual information so long as the generalization can capture features that determine what kind of generalization it is. So a cat's ear is basically triangular but there are details that differ from cat to cat, so the details can differ so long as your process can capture the generalization of "triangular". Now that's just an example, the problem becomes as to what level of resolution should your generalization break the entire visual universe into? One approach is to use fractals and break everything into such geometry. 

[BayramHm]

Thanks a lot bro, you mean i use fractals for the two images and compare them ?

[frankinstien]

Thanks a lot bro, you mean i use fractals for the two images and compare them ?

Basically yes.