File talk:Classification of images progress human.png

While this graph is very useful, I question the inclusion of the 5% line for humans.

It's clear that for humans this is caused by the testers' lack of awareness of the difference between species and similar minor differences in the dataset. Whereas the machines' errors are typically gross mistakes.

The supporting papers make these caveats clear but this will not be obvious to the typical Wikipedia reader who will get the impression that these algorithms are better overall than humans.

The graph would better serve a NPOV presentation by removing the red line. Chris55 (留言) 23:48, 18 December 2016 (UTC)