At lower left we see the ommatidial lenses themselves, shifted into focus by the overall curvature of the compound eye.

The above is 100% crop of a single image, Mitutoyo 20X M Plan Apo objective with Raynox DCR 150 tube lens, on Canon T1i sensor (pixel pitch 4.69 microns).
Following are some images that may help to make more sense of what's being shown.
First, the overall setup. This is just a shed skin that happened to end up with the compound eye corneas oriented almost horizontally, positioned over a chrome-on-glass USAF resolution chart, illuminated from below by a small flashlight bounced off an index card. We're concerned with the stuff inside the green rectangle.

Here is the whole frame, as seen through a 5X objective, focused on the resolution chart.

Again the whole frame at 5X, but now focused on the ommatidia images. This focal plane is 4.395 mm higher than the resolution chart.

Here, the whole frame at 20X. The first image shown in this thread is a 100% crop from this frame.

Comparing the resolution chart seen directly, versus through the ommatidia lenses, it's clear that we're looking at real inverted images formed behind the corneal lenses.
Comparing the size of the resolution chart bars as seen directly and through the ommatidia lenses, the magnification of the ommatidia lenses calculates out to be 1/113 X. Then from the 4.395 mm separation of focal planes and the magnification, the effective focal length of the ommatidia lenses calculates to be about 38 microns. The separation between adjacent ommatidia is about 27 microns, so allowing for the seams, these lenses are about f/1.8 .
I give thanks to Giorgio Pattarini (username patta) for our discussion in https://www.photomacrography.net/forum/ ... 27&t=42758, which prompted me to look again at this specimen. My earlier impression of it -- formed too quickly without looking closely -- was totally incorrect.
--Rik