With the help of Dr. Yampolskiy, I put the information in the series on optical illusions together into an academic paper, which can be found here: https://arxiv.org/abs/1810.00415

The main goal of creating the dataset is more or less complete, but as it stands it is not super useful and could use some reorganization.

I sent the paper to Luke Miles to see what he thought (after submitting to Arxiv, unfortunately) and he gave the following feedback:

Unrequested feedback on Max’s paper:

• That was actually a good read
• The 1 footnote on Max’s name is never resolved
• First couple sentences of abstract are not relevant
• Backwards quotes?
• Italic author names?
• Is confusion matrix test or train?
• Column labels on confusion matrix?
• For consistency, section 4.2 should be called “Generative Adversarial Network Results”
• In section 5, clarify that the vision system of interest is the eyes of predators. Even if butterflies couldn’t see, they would evolve fake eyes to fool bluejays. Feedback is via the killing of butterflies (by owls) who wear weak illusions.
• Good choice to host the images on floydhub; I had like 8MB/S download speed. Maybe clarify that you can download them for free without an account, because floydhub may try to get them to sign up first (like on dropbox)

I didn’t realize latex required specifying quote direction, so all of my quotes are backwards! How embarrassing. I won’t re-upload for cosmetic changes because that’s against Arxiv’s archiving philosophy, but when I do make another version I will incorporate these changes.