kid's gender and age detection is not accurate

[pengKiina]

running many images, every times comes to kids faces, the gender and age's detection is far from accuracy.

Is there some way can get better accuracy?

Thanks!

image link https://unsplash.com/photos/oWDRVgk04EA

from deepface import DeepFace
DeepFace.analyze((images[1]))

{'emotion': {'angry': 0.04058407163003514,
  'disgust': 5.642185721868631e-06,
  'fear': 58.03913820023168,
  'happy': 0.0008088882650657082,
  'sad': 0.0009563027221449876,
  'surprise': 41.91847991457462,
  'neutral': 3.256737296336301e-05},
 'dominant_emotion': 'fear',
 **'age': 29**.03933189735305,
 **'gender': 'Man',**
 'race': {'asian': 2.548585238968446,
  'indian': 1.3059119540996524,
  'black': 0.28957356096497544,
  'white': 57.22050411823417,
  'middle eastern': 14.436330761690781,
  'latino hispanic': 24.19909234041526},
 'dominant_race': 'white'}

[serengil]

Overall, age prediction has almost ±4 year error. This is acceptable because you predict someone in which decade group.

Besides, predicting children underperform because the source data set (imdb + wiki) has limited children images. Here you can find the distributions: https://sefiks.com/wp-content/uploads/2019/02/age-gender-distribution.png

We can find a larger data set in the future and re-train the model.

This is not a bug, that's why, I'm going to close this issue.

Tutorial: https://sefiks.com/2019/02/13/apparent-age-and-gender-prediction-in-keras/

[pengKiina]

I see, thanks a lot.