So for example in the first test (top left collection 3x2 of six total pictures), the second picture in the first row has what I'll call a squashed cup icon with whipped cream on top in two different orientations. So I'd click on that one.
The for the second test (second collection of 6 pictures in the top row), the very first picture has two icons of a "dollar bill" in two different orientations. So I'm assuming you'd select that one.
For the third test (third/final collection of 6 pictures in the top row) the second picture in the first row has two icons of a briefcase in two different orientations.
While I didn't complete all 12 "tests", each one I looked at seemd to have one of the 6 pictures having a duplicate icon.
But yes definitely harder than the: "Does this image have a cross walk?" Or "Does this image have a lamppost?" Which I assume are pretty easily ML solvable now.
At least share the answer ?