Support for automatic image classification and tagging via something like mobilenet.
Questions for y’all:
The tag quality can be confusing/wrong/racist. Should those terms only be exposed via search results? Via the info panel? Via current tag browsing?
If I support browsing these terms, should they be added to the Keywords/ hierarchy, or some place else (as there will be hundreds-to-thousands of terms added to libraries).
Haven’t tried PhotoStructure yet, but I’d love this feature and have a few thoughts:
#1, on one hand, it would be nice if tags could be exposed so that manual corrections can be made. I see two patterns: a) I’m viewing an image, notice a tag is wrong, and do a correction on a spot-basis. b) I want to do a correction en-mass of a problematic tag (e.g. search on the tag, select a bunch of images to remove it from). Any manual edits I make should override the generated tags, and persist if an updated version comes along with a better classifier that wants to rebuild my image tags (unless I explicitly elect to clear everything and start from scratch).
I for one understand you are not Google and would set my expectations regarding accuracy accordingly. And since you aren’t Google and the data would stay local, I’d be forgiving of embarrassing or offensive false positives. I know it won’t be perfect out of the box, but there should be a pathway to keep up as the field advances.
#2, on the other hand, if keywords would already be an effective way to accomplish some of the overrides I’ve envisioned above, I’d probably be OK if the AI tags were kept separate and even static (non-editable). Guessing that would simplify implementation. I’d want to be able to search and filter using both tags and keywords interchangeably.
Hope this is constructive, and not too marred by my lack of familiarity (so far) with the software.
I suggest putting under Keywords/AutomaticTagging so you can distinguish them from the manual keywords. And also writing those keywords back to EXIF metadata
I tried it, and it’s garbage. Some people may have a use for it, but if I desperately need to see a picture of a banana - I can use google image search. In my opinion, face tagging would be 1,000X more useful than generic image tagging.
Also, the most horrifying part of Plex’s implementation is they send all of your pictures to a 3rd party for tagging.