On 2017 Mar 11, Lydia Maniatis commented:

Reading this article, one gets the impression that the authors don’t quite believe their own claims, or aren’t really sure about what they’re claiming. This is illustrated by the following statement (caps mine): “It could be that other factors often associated with perceptual inferences and top-down processing—scission, grouping, object formation, and so forth—could affect the filters OR ACTUALLY ARE THE FILTERS” (page 17).

Whereas up to this point the notion of “filters,” described as acting “early” in the visual process, has referred to a technical manipulation of the image based on sharpening or blurring luminance “edges” (a process which is unnecessarily and somewhat confusingly described in terms of removing “low or high spatial frequency content,” even though the elements referred to are not repeating) we are now told that this manipulation - the “simple filter” - may be equivalent to processes of perceptual organization that produce de facto inferences about the distal stimulus, such as the process of figure-ground segregation (which is, of course, prior to “object formation”). This is quite a surprise – in this case we perhaps could refer to, and more explicitly address, these organizing processes - assuming the authors decide definitively that this is what they mean, or unless the term “filter” is only intended to mean “whatever processes are responsible for certain products of perception.”

With respect to the specific story being told: As with all "spatial filtering" accounts to date, it is acknowledged to be ad hoc, and to apply to a small, arbitrary set of cases (those for which it has been found to "work"). Which means that it is falsified by those cases which it cannot explain. The ad hoc-ness is incorporated into the “hypothesis: “The results support the hypothesis that, under some conditions, high spatial frequency content remains invariant to changes in illuminant” (p. 17). Which conditions are being referred to is not specified, begging the question of the underlying rationale. The authors continue to say that “Of course, this hypothesis may not be true for complex scenes with multiple illuminants or large amounts of interreflection.” Arguably, all natural scenes are effectively under multiple illuminants due to shadows created by obstructions and orientations relative to light sources.

In fact, as with all filtering accounts to date, the account doesn’t even adequately explain the cases it is supposed to explain. The reason for this is that it doesn’t address the “double layers” present in perception with respect to illumination. It isn't fair to say that when a perceived surface is covered by a perceived shadow, we are discarding the illumination; the shadow is part of the percept. So to the extent that Dixon and Shapiro’s manipulation describes the perceptual product as containing only perceived surface lightness/color values but not illumination values, it is not representative of the percept corresponding to the image being manipulated.

Relatedly, Dixon and Shapiro don’t seem to understand the nature of the problem they are addressing. They say that: “Most explanations of the dress assume that a central task of color perception is to infer the reflectance of surface material by way of discounting the illumination falling on the object” (p. 14). This may be accurate with respect to “most explanations” but, again, such explanations are inapt. As I have noted in connection to one such explanation (https://pubpeer.com/publications/17A22CF96405DA0181E677D42CC49E), attributing perceived surface color to perceived illumination is equivalent to attributing perceived illumination color/quality to perceived surface color. They are simultaneous and correlated inferences, and treating one as a cause of the other is like treating the height of one side of a seesaw as the cause of the height of the other side. You need to explain both what you see and what you saw.

The confusion is similarly illustrated in Dixon and Shapiro's description of Purves’ cube demo, described as “an iconic image for illustrating the effect of illumination on color appearance…” (p. 3). But the demo is actually not illuminated if on a computer screen, and if observed on a page, is typically observed under ordinary lighting. Again, both the color of the surfaces and the color of the illumination are inferred on the basis of the chromatic structure of the unitary image (of its retinal projection); and both are effects, not causes, and both are represented in perception. I see the fabric of the dress as white and gold, and the illumination as bluish shadow. Neither is “discounted” in the sense that Dixon and Shapiro seem to be claiming.

With respect to the problem of the dress specifically, none of these explanations address why it interpreted one way by some, another way by others. The ambiguity of light/surface applies to all images, so general explanations in terms of illumination/surface color estimation don't differentiate cases in which there is agreement from those rarer ones for which there is disagreement. The reference to one perceptual outcome of the dress as indicating “poor color constancy” or “good color constancy” is inapt as the images do not differ in illumination, or surface color, but only in interpretation.

As I've also noted previously, the proof of understanding the dress is to be able to construct other images with similar properties. So far they've only been found by chance.

On 2017 Mar 11, Lydia Maniatis commented:

Reading this article, one gets the impression that the authors don’t quite believe their own claims, or aren’t really sure about what they’re claiming. This is illustrated by the following statement (caps mine): “It could be that other factors often associated with perceptual inferences and top-down processing—scission, grouping, object formation, and so forth—could affect the filters OR ACTUALLY ARE THE FILTERS” (page 17).

Whereas up to this point the notion of “filters,” described as acting “early” in the visual process, has referred to a technical manipulation of the image based on sharpening or blurring luminance “edges” (a process which is unnecessarily and somewhat confusingly described in terms of removing “low or high spatial frequency content,” even though the elements referred to are not repeating) we are now told that this manipulation - the “simple filter” - may be equivalent to processes of perceptual organization that produce de facto inferences about the distal stimulus, such as the process of figure-ground segregation (which is, of course, prior to “object formation”). This is quite a surprise – in this case we perhaps could refer to, and more explicitly address, these organizing processes - assuming the authors decide definitively that this is what they mean, or unless the term “filter” is only intended to mean “whatever processes are responsible for certain products of perception.”

With respect to the specific story being told: As with all "spatial filtering" accounts to date, it is acknowledged to be ad hoc, and to apply to a small, arbitrary set of cases (those for which it has been found to "work"). Which means that it is falsified by those cases which it cannot explain. The ad hoc-ness is incorporated into the “hypothesis: “The results support the hypothesis that, under some conditions, high spatial frequency content remains invariant to changes in illuminant” (p. 17). Which conditions are being referred to is not specified, begging the question of the underlying rationale. The authors continue to say that “Of course, this hypothesis may not be true for complex scenes with multiple illuminants or large amounts of interreflection.” Arguably, all natural scenes are effectively under multiple illuminants due to shadows created by obstructions and orientations relative to light sources.

In fact, as with all filtering accounts to date, the account doesn’t even adequately explain the cases it is supposed to explain. The reason for this is that it doesn’t address the “double layers” present in perception with respect to illumination. It isn't fair to say that when a perceived surface is covered by a perceived shadow, we are discarding the illumination; the shadow is part of the percept. So to the extent that Dixon and Shapiro’s manipulation describes the perceptual product as containing only perceived surface lightness/color values but not illumination values, it is not representative of the percept corresponding to the image being manipulated.

Relatedly, Dixon and Shapiro don’t seem to understand the nature of the problem they are addressing. They say that: “Most explanations of the dress assume that a central task of color perception is to infer the reflectance of surface material by way of discounting the illumination falling on the object” (p. 14). This may be accurate with respect to “most explanations” but, again, such explanations are inapt. As I have noted in connection to one such explanation (https://pubpeer.com/publications/17A22CF96405DA0181E677D42CC49E), attributing perceived surface color to perceived illumination is equivalent to attributing perceived illumination color/quality to perceived surface color. They are simultaneous and correlated inferences, and treating one as a cause of the other is like treating the height of one side of a seesaw as the cause of the height of the other side. You need to explain both what you see and what you saw.

The confusion is similarly illustrated in Dixon and Shapiro's description of Purves’ cube demo, described as “an iconic image for illustrating the effect of illumination on color appearance…” (p. 3). But the demo is actually not illuminated if on a computer screen, and if observed on a page, is typically observed under ordinary lighting. Again, both the color of the surfaces and the color of the illumination are inferred on the basis of the chromatic structure of the unitary image (of its retinal projection); and both are effects, not causes, and both are represented in perception. I see the fabric of the dress as white and gold, and the illumination as bluish shadow. Neither is “discounted” in the sense that Dixon and Shapiro seem to be claiming.

With respect to the problem of the dress specifically, none of these explanations address why it interpreted one way by some, another way by others. The ambiguity of light/surface applies to all images, so general explanations in terms of illumination/surface color estimation don't differentiate cases in which there is agreement from those rarer ones for which there is disagreement. The reference to one perceptual outcome of the dress as indicating “poor color constancy” or “good color constancy” is inapt as the images do not differ in illumination, or surface color, but only in interpretation.

As I've also noted previously, the proof of understanding the dress is to be able to construct other images with similar properties. So far they've only been found by chance.