Using IBEIS for Image Visualization

[Erotemic]

This post is in response to an email I received:

We’re investigating various visualization techniques, such as XAI and image registration methods, to help end-users analyze predicted matches.
We’re interested in using Hotspotter to generate graphics for pairs of images. We don't plan to use HotSpotter for determining potential matches, just for producing visualizations that highlight salient features across images. Could you advise on a way to leverage the Hotspotter code for this purpose?

The visualization code I've used I think is fairly separable from the rest of the system.
In the ibeis repo there is a submodule called ibeis.viz which is the logic that controls all IBEIS visualizations: https://github.com/Erotemic/ibeis/tree/main/ibeis/viz
This will depend heavily on the code in plottool_ibeis module: https://github.com/Erotemic/plottool_ibeis 
The code to generate the heatmaps based on matching elliptical keypoints lives in vtool_ibeis because it draws on images directly rather than using matplotlib (which is what plotool_ibeis is mostly for): https://github.com/Erotemic/vtool_ibeis/blob/main/vtool_ibeis/coverage_kpts.py
Some of the last code I wrote when I worked full time on the project was a GUI for inspecting 1-vs-1 matches in an interactive way where I could change the parameters of the keypoint+descriptor extraction and matching algorithm to quickly and interactively see the results. This code also lives in vtool_ibeis: https://github.com/Erotemic/vtool_ibeis/blob/main/vtool_ibeis/inspect_matches.py
Much of the referenced code should have doctests, which provide a minimal entry point into the code. Some of them will be decorated with a comment like # SCRIPT or # xdoctest: +SKIP or # DISABLE_DOCTEST, which means that code isn't run in GithubActions, so I can't guarantee that it will still work, but I believe much of it will. 
For instance, I just ran pip install ibeis on my machine, and ran one of the doctests for the MatchInspector class in vtool_ibeis.inspect_matches: 

        >>> from vtool_ibeis.inspect_matches import *  # NOQA
        >>> import vtool_ibeis as vt
        >>> gt.ensure_qapp()
        >>> ut.qtensure()
        >>> annot1 = lazy_test_annot('tsukuba_r')
        >>> annot2 = lazy_test_annot('tsukuba_l')
        >>> match = vt.PairwiseMatch(annot1, annot2)
        >>> self = MatchInspector(match=match)
        >>> self.show()
        >>> # xdoctest: +REQUIRES(--show)
        >>> #self.update()
        >>> gt.qtapp_loop(qwin=self, freq=10)

Which still seems to work and popped up the matching GUI: 

Toggling the lines and ellipses in the display config gives:

You should be able to reuse/repurpose visualization code with any sort of ellipse base keypoints and descriptors (or even just a homography warping between two images).