gallery tab: similarity search

[picobyte]

commit 34be00c added a gallery tab.

Currently it is only for displaying duplicates. I intend to also use it for a rudimentary image browser, but that should be handled in another issue. This issue is to discuss image similarity.

I've done some tests, and in my test hash based (file rename), a slight crop, horizontal flipped image, scaled down image, they were all reported as duplicates, which was a surprisingly quick success.

Duplicates now are based on both checksum and sorted tags, including the weights, truncated precision (one decimal).

• The hash and tags based duplicate search can cause duplicates of duplicates, currently. which should be fixed.
• The unfiltered tags dict is currently used, but maybe the filtered tags could be used. This would allow for a similarity of selected features search. Maybe the search field could be included. I intend to make this a multi search, comma based, but again, another issue.
• There could be a tag weights based sorting of (duplicate/similar) images.
• currently if there are three files duplicate of one another, that is not shown, though they are listed subsequently.
• There should be a text display of resolution and filename (basename, including extension, file size and maybe date) also it could be displayed based on what this was a duplicate.
• It should be possible to delete one of the images or move images elsewhere.
• An auto-select for deletion may also be useful. for inspiration on how to do this see my bash script tag_based_image_dedup.sh

Already an image can be dragged in the single image interrogation window. Maybe this should trigger a tab switch and subsequent image interrogation.

I believe an image similarity search may be something that could be useful, to not only display the exact duplicates, but also images that are very similar, to avoid over-training. Let me know if you have something to add, I'm fairly naive still regarding the training based on images.

Also please note that which interrogator you pick can have an effect on the duplicates that you may find. This could be interesting for people who are create these models for tagging.

[sdbds]

we can Refer to this.

[picobyte]

Interesting, I was implementing a similarity search within the batch of interrogated images that you have tagged.