Why so many false positives?

Question

I tested this API and it seems to consider every image containing a woman a NFSW.

https://thumbs.gfycat.com/GroundedThatEuropeanfiresalamander-mobile.jpg - 89% confidence, where I would expect around 60%, because you can see her shoulders so this is a tricky image (still no nudity)
https://i.vgy.me/Kd8Aik.png - 99%, is this because of open mouth?
https://i.vgy.me/5mxSQe.png - 94%, even more than first image - why?

It looks like this model was trained on porn images only - it doesn't know anything about NOT porn?

https://i.vgy.me/a2ntyh.png - single color image - 13%
https://i.vgy.me/K4s9Ds.jpg - at least the nature got really low score of 1%

litmind · Answer

Same here, tested with thousands of manually classified photos and this model is not trained to properly detect NSFW images, and cannot be trusted to do so.

NSFW Image Classification

9.6 / 10

1,293ms

100%

N/A

Why so many false positives?