Created attachment 169469 [details] Examples for new features I would like to suggest two features making the face-recognition workflow easier, faster and more precise to use especially with bigger photo and face libaries 1. Make it possible to clear all unconfirmed face suggestions without opening "unconfirmed" tag Reasoning: Running a bigger face-recognition task leaves me with thousands of unconfirmed faces. Cleaning them up requires me to open the "unconfirmed" tag, oftentimes taking multiple minutes to open or even crashing the application. I guess it has to do with the face-thumbnails not being lazyloaded. 2. Add a filter for the face recognition limiting the facetags used in the workflow. (just as it is possible to limit the process to certain albums or tags). Reasoning: The bigger the photo library and the more facetags exist in my database, the longer take simple actions (e.g. importing a single album and scanning for faces) as the comparison has to be made for every single facetag with every image. On top of that, the results for e.g. birthdayphotos with my grandma are scattered around all the possible facetags even though I know there is a limited amount of people possible on these pictures. I end up going through my entire facelibrary to get them all together and assigned correctly. Therefore it would be easier to limit the possible results of the face-recognition task to certain tags (equally as it possible to limit the process to certain albums). That would also make it easier and faster searching for a single facetag in a bigger range of albums. This would probably also improve the perceived accuracy of the recognition-workflow.
Point 1 is a duplicate of Bug 435146 and Bug 449456 Point 2 A type of filter to avoid displaying all unconfirmed faces is to select the person who has unconfirmed faces, unconfirmed faces have a green frame. The people are sorted upwards depending on the number of unconfirmed faces. Maik
Re Point 2: That's not what I meant. I want to be able to apply the filter before running the face-recognition, not afterwards. So that the face suggestions are only associated to the previously selected facetags (see attachment).
Michael, With this entry, only the point 2/ must be considerated. Point 1/ is already reported in another files. Gilles
(In reply to roqstr from comment #2) > Re Point 2: That's not what I meant. I want to be able to apply the filter > before running the face-recognition, not afterwards. So that the face > suggestions are only associated to the previously selected facetags (see > attachment). Hi rogstr, First, thank you for taking the time to submit a request. It's people like you who really help make digiKam better. Unfortunately, we can't limit searching by existing tags. We search for face matches using an in-memory structure that holds all the "signatures" of existing confirmed faces. The benefit is that searching is very efficient and very, very fast. The most time-consuming part of adding a new image and searching for faces has been the face detector and the extractor. The extractor is what generates the signature used for matching. In digiKam 8.5.0 we're introducing a new face detector called YuNet. YuNet is between 6 and 20 times faster than YOLOv3, depending on settings. I think this change will help speed things up considerably. Also, the new extractor/recognizer, called SFace, is much more accurate than OpenFace. This will significantly reduce false-positive matches, so we don't have limit the confirmed faces we're searching against. I hope you'll try digiKam 8.5.0 and let us know if think there are improvements. You can try a daily build of digiKam, but we don't recommend you use for production. Cheers, Mike
Hey Mike, thank you for your in-depth reply. I really appreciate the work you put into the face detection and recognition this release. I've tried it out today and it makes a giant difference. In my opinion the suggested feature is obsolete the changes coming in 8.5 and this ticket can be closed.
Hi rogstr, Thank you for the reply. It's great news you like the improvements we've made in 8.5.0. Keep the suggestions coming! Cheers, Mike