| Summary: | Automatic face recognition gives up on some faces | ||
|---|---|---|---|
| Product: | [Applications] digikam | Reporter: | HubMiner <cal030> |
| Component: | Faces-Recognition | Assignee: | Digikam Developers <digikam-bugs-null> |
| Status: | RESOLVED NOT A BUG | ||
| Severity: | normal | CC: | caulier.gilles, michael_miller |
| Priority: | NOR | ||
| Version First Reported In: | 8.6.0 | ||
| Target Milestone: | --- | ||
| Platform: | Microsoft Windows | ||
| OS: | Microsoft Windows | ||
| Latest Commit: | Version Fixed/Implemented In: | 8.7.0 | |
| Sentry Crash Report: | |||
|
Description
HubMiner
2025-04-17 11:41:42 UTC
(In reply to HubMiner from comment #0) > After training and confirming People tags (20K assigned, 14K ignored), I > have a few hundred candidates in Unknown. Some are very clear and are a > good representation of that person. After a few training cycles, the face > AI stops suggesting any more face matches. > > Extending that scenario - if there is a group of similar faces in the > Unknown set, and I manually assign one or couple faces, and rerun Recognize > faces, sometimes it will trigger to recognize the rest of that group, but > sometimes it will not. > > An additional idea: Although AIs are typically black boxes, it would be > helpful to have some feedback from the tool why it does not propose > assignments. Hi HubMiner, The face engine is comprised of 3 main parts: face detection, feature extraction, and classification (matching). If you are seeing the face in the People->Unknown view, then the first step was successful. digiKam may not match a face for 2 reasons. The first reason is the AI can't extract the features, which means it can see that it is a face, but it can't get the details it needs to create a "thumbprint" of the face. The second reason is that digiKam can't or won't match the "thumbprint" to other faces. You can easily test the second scenario by setting the recognition accuracy slider to something low, like 2 or 3, and then doing a recognition scan. Don't worry, this won't mess up your library. You can put it back to exactly where it is by setting the accuracy slider back to it's present value and then doing another recognition scan. Try that and see if the face get's matched to anyone. Cheers, Mike The Face Detection was at 10, when dropped it 7 is made more good matches. Below that, it kept making the same wrong matches. Maybe we need some way to hint to a user that this setting has an impact. I think we can close this issue at this point. Thank you for your hard work! (In reply to HubMiner from comment #2) > The Face Detection was at 10, when dropped it 7 is made more good matches. > Below that, it kept making the same wrong matches. > > Maybe we need some way to hint to a user that this setting has an impact. > > I think we can close this issue at this point. > Thank you for your hard work! Hi HubMiner, The default settings for both sliders in the face settings is 7. The digiKam documentation states the default settings have been carefully chosen based on thousands of tests. Also, if you hover your mouse over the slider you'll see a tooltip explaining how they affect accuracy. Cheers, Mike |