Perfect setup for grouping children’s faces with Face Recognition

🍱 Features & apps
1

Hello!
I am struggling a bit by using the app „Face Recognition“ to correctly group baby’s and young children’s faces (~ younger than 6). I tried using model 1 and 4 … for both I am getting young people grouped quite well with a clustering threshold of max. 0.31 – but this causes too many groups to assign for adults. The other way around, if I increase the threshold to e.g. 0.5, I am getting very good results for adults, but all children appear in only one group.

I was also experimenting with the minimum confidence. But without success so far.

Does anyone maybe figured out how to deal with children’s faces in „Face Recognition“? What is the perfect value setup to get good results for both people groups – young & adults?

I also remember using “Recognize” some time ago with a similar result. Maybe it’s handling young people’s faces better, now!?

Thanks a lot for sharing your experiences.

2

also interested in people’s experience about this…

3

@alexanderdd : It seems that the model behind facerecognition is simply not trained for younger faces. I found this message by the author on github:

The model was trained with photographs of middle-aged people, Therefore, it just can’t identify them correctly. That is, the model is blinded, and all the boys appear to be the same. Unfortunately, the same applies to any non-Caucasian persons, and these are simply all the same for the model… :disappointed:

Source: Feature idea: age-aware recognition · Issue #620 · matiasdelellis/facerecognition · GitHub

I guess it’s the same with the models used in Recognize in the end.

1 Like