Bug 513269

Please provide a debug log from the beginning of the digiKam startup. I suspect a problem with the face database.
No training data is available.

For an SQLite database, it's best to delete the "recognition.db" file and rebuild it using the maintenance tool.

Maik

Created attachment 187565 [details]
startuplog

Here's the log from the startup of digikam. I removed the recognition.db upfront, and could see the db was recreated, but not touched (just the same 32kB initial size).
Tried maintenance->retrain and some manual tagging.

All your faces in the images are being filtered out because they are too small. This behavior is intentional; to use only good samples as training data, small and blurry faces are filtered out.

Without a sample image, it's difficult to verify if this is the case.

FacePipelineBase::useForTraining Image is too small for face detection -2147483648 x -2147483648 required: 44.8
FacePipelineEdit::writer(): not using for training:  "F:/...031225/20191222_192026.jpg"

Maik

Created attachment 187573 [details]
startuplog_raw

The log file was a bit misleading unfortunately because I stripped away a lot of log lines to make the file smaller and "easier" to read (I thought). You can see a jump from serial nr5 to 1899 of files which are not too small. It's just around 82 which are too small, and the rest of the 1907 seven pictures are normal size with normal people :)
I've attached the raw log file now. I'll attach an example image as well.

Here's from the old file with jump from 5 ->1899
00016855	229.29008484	[3736] digikam.facesengine: FacePipelinePackageBase::~FacePipelinePackageBase: Deleting package with serial number 5	
00016856	229.33230591	[3736] digikam.facesengine: Finish aligning face in  11  ms	

00024426	277.08789063	[3736] digikam.facesengine: Start neural network	
00024427	277.09997559	[3736] digikam.facesengine: Finish computing face embedding in  12  ms	
00024428	277.10046387	[3736] digikam.facesengine: FacePipelinePackageBase::~FacePipelinePackageBase: Deleting package with serial number 1899

Created attachment 187574 [details]
samplepicture

Git commit f8137ca37c994831e5cc1f1e39d26df4c3ee98f4 by Gilles Caulier.
Committed on 13/12/2025 at 12:51.
Pushed by cgilles into branch 'master'.

add explaination about low quality images ignored while face workflow

M  +1    -0    left_sidebar/people_view.rst

https://invent.kde.org/documentation/digikam-doc/-/commit/f8137ca37c994831e5cc1f1e39d26df4c3ee98f4

Created attachment 187686 [details]
retry_facetaggin

I moved the pictures it warned about in previous logfiles outside of digikam workfolders (even though these pictures are typical 3024*4032 and around 3,5-4 MB, I'll attach a screenshot of these "to small" as well)

I did a maintenance->rebuild training data, and then people-> recognice faces only.  Then started tagging people. The only thing happening is it starts a background process which restarts on every new tagging. Attached the log from this session.

Created attachment 187687 [details]
tosmall_warning_listview

A listing of the pictures which were giving the following warning (the preview picture is among this warning here):

00001233	96.17277527	[3736] digikam.facesengine: FacePipelineBase::useForTraining Image is too small for face detection -2147483648 x -2147483648 required: 44.8	
00001234	96.17285919	[3736] digikam.general: Found FacesEngine identity 1 for tag 25	
00001235	96.17289734	[3736] digikam.facesengine: FacePipelineEdit::writer(): not using for training:  "F:/... 031225/20200303_172256.jpg"	
00001236	96.17320251	[3736] digikam.facesengine: FacePipelinePackageBase::~FacePipelinePackageBase: Deleting package with serial number 38	
00001237	96.18213654	[3736] digikam.facesengine: Finish aligning face in  9  ms