I briefly thought this might be a discussion of Flat Earthers.

Take a look at the individual segments zoomed in: https://i.imgur.com/H4y4XTH.png

I think what's happening is the brain is processing the separate color segments individually, instead of a continuous line. And at an individual segment level, it's much harder to detect a curve on the bottom pieces than the top. Notice the red line I drew. The bottom pieces fit much closer to a true line than the top pieces.

You can play with the colors here: https://www.geogebra.org/m/MTGuYMTC

This is a very interesting visualization. If you adjust col2, you can see that the illusion basically has a step function. If both line colors are darker than the background, the two patterns look the same. If you slowly adjust col2 until it matches the background and then becomes just a hair lighter, they pop into existence with the full illusion. There is no transition whatsoever except insofar as you can't see the lines when they match the background color.

How are you adjusting the colours? The page is not interactive on my device, and I see no controls of any sort.

<Insert BS misconception of the illusion, and why you know better than the researchers> here.

"It's just image compression artifact" took the crown for me.

the zig-zags where the color change is at the peaks / troughs looks to me like a wall with light coming from the right side. i wonder if that's what's going on here, the brain is saying "i know that shape, it's a zig-zagging wall"

This seems not to be due to curvature but due to color distribution of lines. I wonder if effect would still be visible if all lines had same color distribution.

It's not supposed to be "due to curvature". It's blindness to curvature, due to the colours.

But it is somewhat dishonest. You could color the lines in the same color as the background and then you would have "illusion" of dissapearing lines, right?

That’s not an illusion. You’d be seeing it accurately.

Now you see why, in much simpler networks, with full observability, it is trivial to generate the equivalent of psychovisual illusions.

Interesting, it's a border dectecion issue, where a single color shows as a curve when it bends sharply the entire way, but short segments approximate to a line when they are mostly just lines with strong curves only at the end.

I suspect it would look very different when animated.

It does indeed look interesting when animated; not sure if this is which parameter you had in mind to vary, but: https://jsfiddle.net/dvv0wd82/

The shift to zig-zag is particularly pronounced when I scan up and down the animation.

In all curves in the image, the shaded area is either centered on the "crown" of the wave or occupies the descending portion of the wave.

What is the visual effect if the ascending portion of each wave is the shaded portion?

You know what this reminds me of? That simple graphical trick to give buttons depth. Two sides have a dark border, two sides have a light border. This seems to be a similar effect.

"However, further experiments revealed that depth perception is not the driving force behind the effect."

I read the paper, but wasn’t immediately convinced it wasn’t related to depth. Once I looked at the illustration for the 3rd experiment [0] it was obviously unrelated to depth.

[0]: https://www.ncbi.nlm.nih.gov/core/lw/2.0/html/tileshop_pmc/t...

I dunno, it seems pretty straightforward to me unless I'm missing something. Color change at the sharpest point of the curve has an antialiasing effect that pronounces the shift and makes it look zigzag.

I wonder if this is connected to the curious zigzags I see in part of my vision field during a visual migraine.

Looks to me like it's because of the average curvature per segment of a constant color.

I don't find this at all "remarkable" because I don't think this has anything to do with the curvature itself, but the relative contrast of the lines against different backgrounds, and the, for lack of a better term, dithering that occurs in the brain as it is interpreting the borders of contrast

If the lines were a solid color, instead of alternating bands, or if the crossover happened somewhere other than a crest of a trough, this effect would disappear.

>I don't find this at all "remarkable" because I don't think this has anything to do with the curvature itself, but the relative contrast of the lines against different backgrounds, and the, for lack of a better term, dithering that occurs in the brain as it is interpreting the borders of contrast

That's because you didn't understand it.

For one, the "the relative contrast of the lines against different backgrounds" is the same between the lines that appear curved and those that appear "straight", because both are the same exact line (both curved) against the same exact background part (the gray).

Correct. The effect is based on where the interchange from black to white occurs. If the interchange occurs at the peak or valley of the sine wave, we perceive no curves. If the interchange occurs at the straight part of the sine way, we accurately perceive the curvature. Really neat!

No, you didn't understand me, sorry. I understood it just fine, but didn't explain myself well enough.

The relative contrast at the point of curvature causes an aliasing effect in the brain where the line colors are (depending on the color) added to or subtracted from the gray value. In effect this causes an apparent sharpening of the curve. However, on those lines where the transition takes place on the straight portion of the wave, no such aliasing takes place because there is no curvature.

It's like anti-aliasing in reverse.

A, fair enough then