I used picasa before (which is a good reference for its UI for face-management!), and like to tag thousands of faces with digikam. Some things I would like to be improved: - The internal face-recognition engine will output some probabilities how similar is a face to an existing face, I guess? At the moment, if I go through the "unknown faces" section, always the last selected Face-Tag is preselected. Perhaps it could prefill with the Tag with the highest probability above a certain threshold? That may speed up things. Also for the drop-down list, sorting by probability. This could be an alternative option, the current behaviour makes also sense in some cases. - When detection is in progress, the "Unknown faces" section is "jumpy", so when I currently are typing on a face-tag-input-field, somehow suddenly another face is selected with the input (this seems like a bug) - it would make sense for the faceview to be able in the preview-pictures-mode to switch between the (currently only available) show-face-only and show-full-picture mode - it would make sense, to have an "and" selection in the faces-tree-view in the left, so if you control-select multiple face-tags, show people together on a picture, which is a very common search I think. I found that is only partially possible via filters (difficult for all images in collection), or the search (also very complicated) - The face-tag-tree-view at the left looks ugly, because the face-preview pictures have different width (it gets worse, if you choose a bigger-preview size like 64px, I use 48px at the moment). At least the face-name should be all same vertical aligned. Also the left margin seems a bit too large (waste of space) - As last point: some few faces have a good recognition (the proposals mostly are correct), but for some faces, almost every proposal is wrong, so there seems to be much room for improvement for the recognition. Also many existing faces are not proposed at all (with accuracy=70%) Thanks for your work at digikam, and for the new face-backend, AI gives a nice boost to such things.
- sometimes it makes sense to rescan the "unknown faces" section, because the engine could be more trained for some faces. But then, already rejected proposals are made again, which is very annoying.
In principle, all existing points have already been reported through existing bug reports. Face recognition only works well with faces detected by digiKam. Picasa faces have a larger cutout and will not work. Are you really using digiKam-7.2.0? In the tags tree, the thumbnails are in the middle, aligned with the thumbnail size, nothing is moved. It looks good to me. Maik
I use the fresh digikam-7.2.0rc appimage. The faces are all recognized by digikam itself. You are right, because the faces have different width, it looks like the names are placed differently, but the are aligned correct. Perhaps it would make sense to display a light grey background+border for the previews to give them equal width? Sorry, that I duplicated some of the reports, and did not split it, but it was so much. I will have a look at existing reports, to check that.
Created attachment 136509 [details] Faces treeview
another possibility: choose a rectangular part of the original image for the preview-treeview if possible (so making the crop a bit bigger), even with the risk, to capture more part of another face.
- when I type a letter into the auto-complete field, it often lists names which do not begin with that letter before names, which begin with that letter, that makes no sense to me.
Well, the search when typing the name is a "contains" search. What if they only know the last name or only part of the name. We did a search with "starts with" and users immediately rejected it as bad. I look at the representation of the face thumbnails again. The distance in the left area is unfortunately also dependent on the widget style and is larger with Fusion than with my native Breeze style. We could remove the "people" parent tag that creates this space. But then face tags can no longer be moved into the main tree using drag & drop. Just take a look... Maik
A contains search seems ok for me, I only meant the order: when I type "T" the list is like: Christoph Martin Thies Tarek It should be more like Tarek Thies Christoph Martin Or better (like said before) adapted by some probabilities, which could be: - face recognition probability - probability, how likely a face appears together with another face (from some collection statistics) - probability, how likely a face appears in a time-span / year / series / folder Such things could really speed up the workflow much