Control number's of face detection on each photos

Scenario:

I have a lot of staged photos. These photos often have a main person (in focus) and other people in the background or next to them (out of focus). I only want to mark the main character with a clear focus (usually only one in a picture).

In the current version (8.6.0), the software will recognize many blurred background faces, which causes a lot of interference in my labeling process.

Possible solution:

YuNet has a top_k parameter to control the result. Can it be changeable by users like the accuracy parameter? By that, I can set the model to only detect one face in every photo.

1 Like

Hi @oxygenkun,
digiKam does not expose most of the parameters passed to YuNet as most users wouldn’t understand them the way you do.

Try setting the Face Size to Extra Large. This will pass a smaller resized image to YuNet making less likely YuNet will discover some of the small faces in the background. It’s not a fix for what you want, but it may help reduce the number of background faces.

Cheers,
Mike

1 Like

Hi Mike,

Thanks for your advice.

I think the option for the maximum limit of face-detection results in each photo is still a quality-of-life feature. The implementation method can also be obtained through the score of the model result, so this option can also be generalized to models that may be replaced in the future.

Best
Ben

1 Like