human level performance on ImageNet, top-1 or top-5?

Question

Anyone have pointers to where the human level performance on ImageNet comes from?

I found a reference to 5.1% accuracy (top-1? or top-5?) from here.

Answer 1

It comes from this paper: https://arxiv.org/abs/1409.0575

O. Russakovsky "ImageNet Large Scale Visual Recognition Challenge" 2014