Ignore vs delete unwanted face annotations

Started by Lord_Helmchen, October 26, 2020, 08:04:00 PM

Previous topic - Next topic

Lord_Helmchen

I have a similar question as in "Ignore Faces vs delete unwanted face annotations?" https://www.photools.com/community/index.php?topic=10415.msg74091#msg74091.

I have started face detection for my DB (210k photos - ca. 100k unique photos). I assigned all faces I want to have in People Manager and ignored unwanted faces (because I like to have the information stored that there is a face). Unfortunately the ignored faces (or better the photos with the faces) still appear in "Unknown Persons" (10,640 files).

My hope was, that the "ignored" faced would not affect performance etc. of IM. But IM gives me a tip for better performance to clean that up via Unconfirmed Faces collection. It says I have 13,577 unconfirmed faces  (but in collections the shown number is 0).

So my question is:
- How can I remove the Unknown Persons (without deleting the Face Annotations)?
- How can I get rid off the "performance info"?

BTW: I got pretty mixed results in face detection: sometimes detection worked really great. But in other cases IM grouped faces and other stuff (e.g. car tires) as same face. But it were totally different faces.

Mario

A) Definition: A face without an assigned person is an "unknown person".
When you set a face to ignore, IMatch no longer tries to automatically assign a person to it. But that does not change the fact that the face has no person assigned, and hence is an unknown person.

B) You cannot 'get rid' of the unconfirmed notice. It is a friendly reminder that your database has 10,000 unconfirmed (!) faces, which IMatch must consider every time you change a person, or new images come in or you change a face assignment etc.

This does not harm much, depending on your database size, computer speed, moon phase etc. It's just a friendly nudge to confirm unconfirmed faces to make IMatch work better.

QuoteBTW: I got pretty mixed results in face detection: sometimes detection worked really great. But in other cases IM grouped faces and other stuff (e.g. car tires) as same face. But it were totally different faces.

Interesting fact: did you know that people say the same thing about the face recognition algorithms from Facebook, Google, Amazon and Microsoft? And that these companies have teams of hundreds of developers, massive amounts of cash and extreme amounts of computing power at their disposal.

I'm just one person and the face recognition rate in IMatch is way over 90% in the typical benchmarks. Which is excellent. There is no 100%.
How you train the AI has a lot of impact. Use the Face Manager to check the faces you have trained and if they are truly representative. Don't train too much. Remove trained faces if they appear only once etc.
-- Mario
IMatch Developer
Forum Administrator
http://www.photools.com  -  Contact & Support - Follow me on 𝕏 - Like photools.com on Facebook

Damit

I have found that ignoring the faces does NOT stop them from said faces being used to try to match to people.  I read the instructions and used ignore to begin with, but I kept seeing the same faces over and over. It was only when I deleted the boxes that IMatch finally stopped trying to group these faces with people for which I had confirmed faces. I went over 30k faces and the fact that ignore did not, in fact, ignore proved to make the experience more frustrating. I kept asking myself "Why am I seeing this face again when I already ignored it multiple times?" This only stopped when I deleted the boxes.

 

Mario

#3
Quoteseeing the same faces over and over.
Meaning?

An ignored face no longer takes part in person assignment.

Do you say that IMatch assigned a person to an ignored face?
Under which conditions?
Where and how did you ignore the face?
Viewer? Face Manager?

AFAIR nothing similar was ever reported. There are no related open bug reports etc.

Please provide detailed information, workflow step-by-step, the IMatch log file in debug mode (see log file) of an IMatch session where you have encountered this unusual behavior.
Maybe a sample set of images which causes this behavior etc.

The more information and details you provide, the more likely it is that I or some other user can reproduce the problem you are reporting.


QuoteI kept asking myself "Why am I seeing this face again when I already ignored it multiple times?" This only stopped when I deleted the boxes.
Deleting a face an ignoring a face is not the same.
-- Mario
IMatch Developer
Forum Administrator
http://www.photools.com  -  Contact & Support - Follow me on 𝕏 - Like photools.com on Facebook

Damit

Yes, I am saying that I used Ignore per the instructions, so that face assignment (face box) would remain but IMatch would not try to assign the face to another individual.

I ignored the faces in Face Manager. I then saw these same faces when I sorted through other people in the face manager. I thought I had misunderstood the function of ignore and then started deleting said faces.  Only then did IMatch stop trying to match these faces to other People. Then I read this thread and realized I did understand the function of ignore faces, but IMatch was not "ignoring" the faces but rather treating these faces like "Rejects," which is my I posted my comment.

Unfortunately, I have sorted through all the 29k unassigned faces so I cannot replicate the behavior or provide the log file as requested. I will import some more folders and test this again and if I encounter the same behavior I will report back here.


Mario

The behavior you see should happen for rejected faces (where you tell IMatch "Not this person") but not for ignored faces.
-- Mario
IMatch Developer
Forum Administrator
http://www.photools.com  -  Contact & Support - Follow me on 𝕏 - Like photools.com on Facebook

Damit

I did some more testing with a smaller group of faces.  It seems that when I Ignored this time around, all those faces went to "Unknown Persons" person in People view.  I assume that is what is supposed to happen. When I looked at them in the viewer, these ignored faces had an X across them, as they should.  Maybe I had so many faces before that they were duplicates? I did not think so, and I did consider it, but maybe I was wrong. 

Anyhow, I will keep an eye on this in the future.

So, when I ignore faces and they go to the unknown persons bin is IMatch's performance impeded at all? I assume since it is no longer trying to match these faces the answer would be no, but I want to confirm this because previously the performance was severely affected during face detection, and I am using a 4 TB WD SN850X for my database. Even now, when I manually add faces in the viewer, I can get long delays will IMatch is "Updating Person." The lag is so long that I have abandoned performing these tasks in IMatch and am now researching how to import face regions and detection from lightroom, which performs better in my experience thus far (but nothing beat the metadata tagging here).

Mario

Face recognition utilizes both all processor cores and the database storage medium to the max.
Performance depends on the processors, their number and how busy your system is.
It is also very important to confirm faces, delete or ignore unwanted faces. The more faces IMatch has to process after each change you make, the longer it will take of course.

Please consider reading the corresponding section in the help: Background Face Processing and the other relevant passages in that help topic.

I run and test the face recognition of IMatch with a database containing 800,000 images and about 85% of these images have one or more faces.
-- Mario
IMatch Developer
Forum Administrator
http://www.photools.com  -  Contact & Support - Follow me on 𝕏 - Like photools.com on Facebook

Damit

I have noticed that this is the first program to actually bring my CPU (Ryzen 5 3600) to 100%.  I was very happy to see a program make use of all the cores! ;D

I have read the section you have pointed out, more than once. That is why I always edited and ignored faces rather than just reject them. I will read it once more. I really try to be proactive and considerate of other's time by really reading and searching the forum for solutions before asking for help.