Bug 384396 - Display faces sorted by similarity (pre-grouped) instead of album/time/..
Summary: Display faces sorted by similarity (pre-grouped) instead of album/time/..
Status: RESOLVED FIXED
Alias: None
Product: digikam
Classification: Applications
Component: Faces-Workflow (show other bugs)
Version: 5.6.0
Platform: Other Linux
: NOR normal
Target Milestone: ---
Assignee: Digikam Developers
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2017-09-05 20:10 UTC by Jens
Modified: 2020-09-12 05:26 UTC (History)
10 users (show)

See Also:
Latest Commit:
Version Fixed In: 7.2.0


Attachments
attachment-25234-0.html (1.50 KB, text/html)
2020-07-04 14:31 UTC, alain.pouchard1410
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Jens 2017-09-05 20:10:47 UTC
When detecting faces, there are often a high number of similar images (and identical faces in different images).
I would like an option where the newly processed images with "unknown" and "unconfirmed" faces are *already* grouped by face so that confirming a face (or creating a new one) is much quicker. 
Correcting such a detection can then be achieved by moving a face from one group to the other.

Note: I do not want the *images* to be grouped, but the *faces*. This means that images with multiple people on them will of course appear in multiple face groups.
Comment 1 Jens 2018-03-18 20:50:37 UTC
Does this wish have a chance of implementation?
I think it is quite useful and many other image management apps implement it (e.g. Apple iPhoto / Photos). When I look for people in images, I don't want to care about files or images, just *people*.
Comment 2 Simon 2018-03-18 20:52:55 UTC
*** Bug 392024 has been marked as a duplicate of this bug. ***
Comment 3 Jmz D3an 2018-05-16 14:10:12 UTC
This would be very helpful when dealing with huge amounts of faces.
Comment 4 stefan.mueller.83 2019-01-27 19:39:48 UTC
I agree.  I regard this a very urgent feature, since if you have a high number of faces, without having a possibility of sorting and filtering, it becomes messy and fiddly what drives me crazy.
Comment 5 Jens 2019-06-19 21:26:27 UTC
Can we please revisit this feature request?
It would help enormously when assigning face tags, even if face recognition is not (yet) perfect.
Thanks!
Comment 6 MarcP 2019-06-20 10:52:22 UTC
In theory, this was proposed as part of the improvements for the GSoC 2019 in the Faces Management workflow improvements project. However, it was included as one of the last things to do (see point 4 at https://docs.google.com/document/d/1EPACR_JgZ5BWfbuqOP_Xlyd_cldEIKSd7n4TTgZ495Q/edit ), so I don't know if that option will be ready this summer.

I agree that it is important, and all other software dealing with picture recognition has this feature.
Comment 7 Maik Qualmann 2019-07-14 13:44:21 UTC
*** Bug 409788 has been marked as a duplicate of this bug. ***
Comment 8 Maik Qualmann 2019-11-30 10:27:54 UTC
*** Bug 414653 has been marked as a duplicate of this bug. ***
Comment 9 caulier.gilles 2019-12-23 15:21:29 UTC
7.0.0-beta1 is out with new Face Recognition algorithm based on Deep
Learning/Neural Network API from OpenCV

https://download.kde.org/unstable/digikam/

Please test and give us a feedback

Thanks in advance
Gilles Caulier
Comment 10 Maik Qualmann 2020-01-04 20:55:43 UTC
*** Bug 415878 has been marked as a duplicate of this bug. ***
Comment 11 Maik Qualmann 2020-02-10 16:52:45 UTC
*** Bug 417390 has been marked as a duplicate of this bug. ***
Comment 12 Jens 2020-02-10 19:20:36 UTC
I looked at 7.0.0beta2.
It seems the new engine better separates unconfirmed and confirmed faces, and also displays unconfirmed faces with the respective people's tag (and a prominent border). Right? However, I cannot find a way to show *only* unconfirmed faces (they are sprinkled between confirmed faces, of which there are many), and I cannot find a way to say "This is a face, but it's not *this* one" to remove a face from the "Unconfirmed" group of faces, but keep the tag.

So far, I also cannot see unknown faces, or images in general, being groupable by similarity. Yet. Is this feature hidden somewhere?
Comment 13 MarcP 2020-02-10 19:37:27 UTC
No, showing only unconfirmed faces for a person (there's current bug report for that), or automatically grouping unconfirmed faces for similarity (this bug report) are features that have not been yet implemented. And rejecting a suggestion still deletes the face region instead of sending that face back to the "Unknown" category (this is bug 384485)
Comment 14 MarcP 2020-02-10 20:09:14 UTC
(I have opened a feature request for the hide/show confirmed faces, so it is registered in the bugtracking system. It is bug 417399)
Comment 15 Kartik 2020-03-19 17:51:57 UTC
Hello all, I would like to work on this bug. 

I noticed that Google Photos resolves this in a very efficient manner, it does not show you the entire "Unknown" Collection, instead only individual collections of people that are currently unknown, but the algorithm has concluded to be the same.

Maybe this is something we can incorporate into DigiKam? I was thinking of creating a hierarchy within the Unknown tag, where each sub-tag will display face-tags of people that are similar (or the same).

The user could reject if the person is not the same as the rest, in which case the person would move into the misc. unknown tag. 

There are a few issues to this, firstly I'm not sure whether it's ideal to create a hierarchy within the unknown tag, furthermore, I'm not sure what I'd name these subtags.

I'd be grateful for any suggestions regarding this.

Thanks 
Kartik
Comment 16 MarcP 2020-03-19 18:00:40 UTC
Hi Kartik. This is quite ambitious! 

Since digikam can already "group" pictures, why not just grouping them by similarity inside Unknown/Unconfirmed, without creating subtags? When the user enters Unknown/Unconfirmed, they will see all the groups (the groups with the most people could be showed first), and then they could enter a name (or confirm the suggestion). The user could also expand or collapse a particular group to see the containing pictures.

Although I believe that the "group" option is disabled for faces.

Another issue will be obtaining the similarity value. I guess that every new face that is detected should be compared with the rest of the faces in the Unknown/Unconfirmed categories and, if the similarity is superior to 0.95 (for instance), then add that face to the group.
Comment 17 Maik Qualmann 2020-03-19 19:07:09 UTC
The digiKam Icon View knows categories. I would find it better if the person name is the category and thus the people are grouped. That would be much clearer.

Maik
Comment 18 Maik Qualmann 2020-05-23 15:51:50 UTC
*** Bug 421985 has been marked as a duplicate of this bug. ***
Comment 19 Kartik 2020-07-03 08:02:24 UTC
(In reply to Maik Qualmann from comment #17)
> The digiKam Icon View knows categories. I would find it better if the person
> name is the category and thus the people are grouped. That would be much
> clearer.

I do like this idea. However, I don't think this would work for images with multiple faces. When categorizing items (faces in our case) the ItemInfo is passed to the various sorting functions. Different faces (in the same image) share the same ItemInfo. Even if I store face suggestions in each ItemInfo, given an ItemInfo it's tough to know which face called it.

I'll have to think of a work-around for this. Maybe I could store a Map of Face Regions to Suggested Faces, and somehow pass Face Regions to the categorizing algorithm as well?

Kartik
Comment 20 Kartik 2020-07-04 12:54:18 UTC
Just to update, storing maps of tag regions to suggested names is definitely feasible. I'll go forward with this approach to implement categorization based on Suggested Names. This will lead to categorization of similar looking faces together.

Kartik
Comment 21 alain.pouchard1410 2020-07-04 14:31:52 UTC
Created attachment 129889 [details]
attachment-25234-0.html

Hello Thank you very much, as soon as there is an online version I will 
test. See you again and again thank you

Le 04/07/2020 à 14:54, Kartik a écrit :
> Just to update, storing maps of tag regions to suggested names is definitely
> feasible. I'll go forward with this approach to implement categorization based
> on Suggested Names. This will lead to categorization of similar looking faces
> together.
>
> Kartik
Comment 22 Kartik 2020-08-01 18:02:29 UTC
Gilles, this bug report may be closed now.
Here's the new Feature in case anyone's interested to see: https://imgur.com/a/CKVopzi
Comment 23 caulier.gilles 2020-09-12 05:26:16 UTC
Fixed with #384396