On 2017 Apr 12, Lydia Maniatis commented:

It occurs to me that the there are two ways to interpret the finding that people were influenced by the stable versions of the dress image in the different settings. The authors say that these versions introduced a bias as to the illumination. But it seems to me more straightforward to assume that they introduced a bias or expectation with respect to the actual colors of the dress, that is a perceptual set mediating the latter. It took me a while to realize this - an example of how explanations that are given to us can create a 'conceptual set.'

On 2017 Feb 15, Lydia Maniatis commented:

I don’t think the authors have addressed the unusual aspects of the dress in saying that “The perceived colors of the dress are due to (implicit) assumptions about the illumination.” As they note themselves, “This is exactly what would be predicted from classical color science…”

The spectrum of light reflected to our eye is a function of the reflectance properties of surfaces and the spectrum of the illumination; both are disambiguated on the basis of implicit assumptions, and both are represented in the percept. The two perceptual features (surface color and illumination) are two sides of the same coin: Just as we can say that seeing a surface as having color x of intensity y is due to assumptions about the color and intensity of the illuminants, so we can say that seeing illumination of color x and intensity y is due to implicit assumptions about the reflectance (how much light they reflect) and the chromaticity (which wavelengths they reflect/absorb) of the viewed surfaces. We haven’t explained anything unless we can explain both things at the same time.

The authors are choosing one side of the perceptual coin – the apparent illumination – and claiming to have explained the other. Again, it’s a truism to say that seeing a patch of the dress as color “x” implies we are seeing it as being under illumination “y,” while perceiving the patch as a different color means perceiving a different illumination. This doesn’t explain what makes the dress unusual - why it produces different color/illumination impressions in different people.

The authors seem to want to take the “experience” route (“prior experiences may influence this perception”); this is logically and empirically untenable, as has been shown and argued innumerable times in the vision literature. For one thing, such a view is circular, since what we see in the first place is a product of the assumptions implicit in the visual process. It’s not as though we see things first, and then adopt assumptions that allow us to see it…In addition, why would such putative experience influence only the dress, and not each and every percept? (The same objection applies to explanations in terms of physiological differences). Again, the question of what makes the dress special is left unaddressed.

It’s odd that, for another example of such a phenomenon, vision researchers need to turn to “poppunkblogger.” If they understood it in principle, then they would be able to construct any number of alternative versions. Even if they could show the perception of the dress to be experience-based (which, again, is highly unlikely to impossible), this would not not help; they would still be at a loss to explain why different people see different versions of one image and not most others. To understand the special power of the dress, they need at a minimum to analyze its structure, not only in terms of color but in terms of shape, which is the primary mediator of all aspects of perception. Invoking “scene interpretation” and “the particular color distributions” are only placeholders for all the things the authors don’t understand.

The construction of images that show that the dress itself can produce consistent percepts is genuinely interesting, but it is a problem that the immediate backgrounds are not the same (e.g. arm placements). This produces confounds. The claim that these confounds are designed to produce the opposite effect of what is seen, based on contrast effects, is not convincing, since the idea that illusions involving transparency/illumination are based on local contrast effects is a claim that is easy to falsify empirically, and has been falsified. So we are dealing with unanalyzed confounds, and one has to wonder how much blind trial and error was involved in generating the images.

Finally, I’m wondering why a cutout of the dress wasn’t also placed against a plain background as a control; what happens in this case? Has this been done yet?

On 2017 Feb 15, Lydia Maniatis commented:

I don’t think the authors have addressed the unusual aspects of the dress in saying that “The perceived colors of the dress are due to (implicit) assumptions about the illumination.” As they note themselves, “This is exactly what would be predicted from classical color science…”

The spectrum of light reflected to our eye is a function of the reflectance properties of surfaces and the spectrum of the illumination; both are disambiguated on the basis of implicit assumptions, and both are represented in the percept. The two perceptual features (surface color and illumination) are two sides of the same coin: Just as we can say that seeing a surface as having color x of intensity y is due to assumptions about the color and intensity of the illuminants, so we can say that seeing illumination of color x and intensity y is due to implicit assumptions about the reflectance (how much light they reflect) and the chromaticity (which wavelengths they reflect/absorb) of the viewed surfaces. We haven’t explained anything unless we can explain both things at the same time.

The authors are choosing one side of the perceptual coin – the apparent illumination – and claiming to have explained the other. Again, it’s a truism to say that seeing a patch of the dress as color “x” implies we are seeing it as being under illumination “y,” while perceiving the patch as a different color means perceiving a different illumination. This doesn’t explain what makes the dress unusual - why it produces different color/illumination impressions in different people.

The authors seem to want to take the “experience” route (“prior experiences may influence this perception”); this is logically and empirically untenable, as has been shown and argued innumerable times in the vision literature. For one thing, such a view is circular, since what we see in the first place is a product of the assumptions implicit in the visual process. It’s not as though we see things first, and then adopt assumptions that allow us to see it…In addition, why would such putative experience influence only the dress, and not each and every percept? (The same objection applies to explanations in terms of physiological differences). Again, the question of what makes the dress special is left unaddressed.

It’s odd that, for another example of such a phenomenon, vision researchers need to turn to “poppunkblogger.” If they understood it in principle, then they would be able to construct any number of alternative versions. Even if they could show the perception of the dress to be experience-based (which, again, is highly unlikely to impossible), this would not not help; they would still be at a loss to explain why different people see different versions of one image and not most others. To understand the special power of the dress, they need at a minimum to analyze its structure, not only in terms of color but in terms of shape, which is the primary mediator of all aspects of perception. Invoking “scene interpretation” and “the particular color distributions” are only placeholders for all the things the authors don’t understand.

The construction of images that show that the dress itself can produce consistent percepts is genuinely interesting, but it is a problem that the immediate backgrounds are not the same (e.g. arm placements). This produces confounds. The claim that these confounds are designed to produce the opposite effect of what is seen, based on contrast effects, is not convincing, since the idea that illusions involving transparency/illumination are based on local contrast effects is a claim that is easy to falsify empirically, and has been falsified. So we are dealing with unanalyzed confounds, and one has to wonder how much blind trial and error was involved in generating the images.

Finally, I’m wondering why a cutout of the dress wasn’t also placed against a plain background as a control; what happens in this case? Has this been done yet?

On 2017 Apr 12, Lydia Maniatis commented:

It occurs to me that the there are two ways to interpret the finding that people were influenced by the stable versions of the dress image in the different settings. The authors say that these versions introduced a bias as to the illumination. But it seems to me more straightforward to assume that they introduced a bias or expectation with respect to the actual colors of the dress, that is a perceptual set mediating the latter. It took me a while to realize this - an example of how explanations that are given to us can create a 'conceptual set.'