Image Classification
Transformers
Safetensors
PyTorch
timm_wrapper
Not-For-All-Audiences

Accuracy improvement opportunity

#2
by DrAliGomaa - opened

Your model is awesome!, just few things noticed after many testing on various types of images, if we can add these types of pics in your dataset that would be appreciated!!

  1. the seethrough images needs improvement a bit(lots of content like this on instagram)
  2. the breastfeeding videos that youtube (allowed for education purposes as they say), the model is doing good in detecting it if the mother is topless, however if she wears something above the breast and allows the baby to feed even if her breast is visible the model don't detect it.
  3. women wearing shorts the model many times don't detect it as low, detected as neutral
  4. Kissing is detected neutral sometimes with a slight medium frequency (0.1-0.2 on average), should be low/medium if romantic kissing.
    Appreciate your work , what you do is meaningful, You are literrally doing a favor to humanity!!, please keep enhancing!.

Sign up or log in to comment