Hacker News Reader

Comment by tgv | Hacker News Reader

Another day, another hyperbolical title. 100 years ago, knowledge of perception was in its infancy. We still don't know much, but we do know that all attempts at a simple, mathematical description of perceptual processes {*} have failed. They're interesting, can serve as an introduction to a problem, may help you to build a better web app, etc., but come nowhere near the complexity of an actual brain. It takes a pretty large neural net to decypher phonemes even in clear speech. Viewing color perception as an independent function mapping one space onto another is already flawed, as the chessboard illusion already shows (https://en.wikipedia.org/wiki/Checker_shadow_illusion).

{*} This is not limited to perception. Attempts to described even small parts of reasoning, language and memory with simple models failed too, but that seems less surprising, since perception is closer to the physical reality.

View on HN

While the title does sound a bit sensationalist the actual paper refers to a standard in the International Commission for Weights and Measures which is indeed (according to the paper) based on 100-year old paradigm introduced by Riemann and furthered by Helmholtz and Schrodinger.

From the abstract:

The scientific community generally agrees on the theory, introduced by Riemann and furthered by Helmholtz and Schrödinger, that perceived color space is not Euclidean but rather, a three-dimensional Riemannian space. We show that the principle of diminishing returns applies to human color perception. This means that large color differences cannot be derived by adding a series of small steps, and therefore, perceptual color space cannot be described by a Riemannian geometry.

https://www.pnas.org/doi/10.1073/pnas.2119753119

This makes sense to me. As I've said elsewhere below, the eye adapts dynamically to each situation. One's perception - the dynamic range of both color and luminance is situational and adaptive.

I know from personal experience of being fooled on hundreds occasions when trying to judge colors, color temperature, etc. despite my experience and the fact that my color vision is normal (I've never failed any Ishihara or related tests).

The law of diminishing returns is essentially what I experience as an observer, the more I grind my observation fine the more fooled and less objective I become. I now only believe my instruments.

Thus it makes sense that large color differences cannot be derived by adding a series of small steps simply for the reason they are in effect a form of noise in relation to large steps.

"therefore, perceptual color space cannot be described by a Riemannian geometry."

Perhaps Riemannian geometry should become just a starting point if for no other reason than it serves to illustrate the complexity of human color perception and the difficulty it poses in trying to analyze it.

>Another day, another hyperbolical title. 100 years ago, knowledge of perception was in its infancy.

That may well be, but the title doesn't point to the overturning of some infantile color theory people believed 100 years ago.

It points to the overturning of a still dominant theory, that was first developed 100 years ago, but continues to be considered as the standard.

The article might be wrong in that that's the case, but that's how the title reads.

Certain models do try to account for spatial localization and relativity too (iCAM), although they are also simple as those effects remain largely unexplored. Honestly, ML seems to be better suited to this task than complex analytical solutions, as long as you have the data to train on.

Reply on news.ycombinator.com