On 2017 Jun 21, Lydia Maniatis commented:

(Comment #2)

The three sentences of the conclusion (which I annotate below, in its entirety, reflects the article's utter lack of content:

1. *"Motion information generated by moving specularities across a surface is used by human observers when judging the bumpiness of 3D shapes." *

The boundaries of specularities are effectively contour lines. It would be thoroughly unrealistic to predict that they would not play a role, moving or not. And the observation had already been made.

1. "In the presence of specular motion, observers tend to not rely on the motion parallax information generated by the matte-textured reflectance component."

The two parts of this sentence seem to be a non-sequitur - how could observers of specular motion employ information generated by matte-textured objects (i.e. objects other than the ones they were observing)? What the authors mean to say is that observers don't use the motion parallax info generated by the specular stimulus. While they frame this as though it were an actual finding, it is, as discussed above, a purely speculative attempt to explain the poorer performance with specular objects.

1. *"This study further highlights how 3D shape, surface material, and object motion interact in dynamic scenes." *

It really doesn't, given the mixed results and failed predictions. It couldn't for a number of other reasons, discussed below.

1. All of the heavy lifting in this article is done by computer programmers, whose renderings are supposed to qualify as "specular" "specular motion" "matte-textured" etc. These renderings rest on theoretical assumptions most of which are never made explicit. They are, however, inadequate; we learn that observers sometimes saw the moving specular stimuli as non-rigid. This is a problem. There is no objective description of the phenomenon "specular object in motion around an axis" other than "objects generated by this particular program." Is there any doubt that results would have been different if the renderings had accurately mimicked the physical phenomenon? If the surface of the object is seen as changing, don't this affect the "motion parallax" hypothesis? The speed with which a particular point on a surface is moving optically is confounded with the speed with which it is moving on its own.

2. The so-called matte-textured objects appeared purely reflective when not in motion. The apparent specularities were "stuck-on" so that they moved with the surface. I have never seen a matte surface with this characteristic. I would be curious to see the in-motion renderings, because I cannot imagine what they look like. What is clear is that a simple reference to matte textured objects is not appropriate. We are talking about a different phenomenon, which may not correspond to any physically actualizable one. This latter fact wouldn't matter if the theoretical framework were tight enough that such stimuli allowed isolation of some particular factor of interest. Here, however, it is just means that "matte" doesn't mean what it normally is thought to mean.

3. Observers were confused about the meaning of the term "bumpiness." Stimuli involve hills and ridges of various extents as well as varying apparent heights. The authors were interested in height. They instructed observers who asked for clarification (not the others) that they were interested in "the amplitude not the frequency." I would say a large hill or a wide ridge could qualify as more ample for people not thinking in terms of graphs with height in the ordinate. In other words, I think there is a observational confound between extent and height of the bumps.

4. In the introduction, the authors refer to previous papers which came to opposite conclusions. Presumably, this means that some relevant factors/confounds were not considered. But the authors don't attempt to analyze these conflicted citations, which thus merely function as window-dressing. They move on to their experiments, on the slightest and vaguest of pretexts, with poorly described stimuli and poorly controlled tasks.

On 2017 Jun 20, Lydia Maniatis commented:

The term "motion parallax" figures prominently as an (oddly negative) explanatory concept in Dovencioglu et al's (2017) brief conclusion. Which is interesting, since we do not come across it in the title, abstract, introduction, methods or results sections of the paper.

It shows up for the first time in the discussion, in the form of a highly speculative suggestion aiming to explain the study's unpredicted and uninterpretable results, in particular why it failed to produce an effect that, throughout the paper, the authors create the impression it did produce.

I don't know if they deliberately meant to mislead, but until I got to a pair of isolated statements in the Discussion, I was under the impression that they were claiming to have shown that "specular motion" (a highly problematic term in its own right, as will be discussed in a subsequent comment) improved observers' ability to make veridical estimates of 3D shape. This is definitely not the case.

While they're careful not to say it outright in the abstract, and while the title, as is typical of vision papers, is uninformative, the prominence of the "specular" term might lead an unsuspecting reader to assume that the claim that results "provide an additional layer of evidence for the capacity of the visual system to exploit image information for shape inference" is referring to some value-added information provided by "specular" as opposed to non-specular, stimuli.

Similarly, the text of the final section of the introduction, titled "Does specular flow facilitate or interfere with 3D shape extimation? " might give the impression that the former is the case: For example:

"...we focus on...whether specular flow...can provide better information on 3D shape than optic flow...specular flows are directly related to 3D curvature and seem to be less sensitive to the particular motion of the object, whereas optic flows vary more substantially with the latter. Thus, if a perceptual task required observers to make judgments abouta an objects' 3D curvature structure...one would expect more consistent shape perception across changes in object rotation axis."

Given the lead-up, I think a reader would be justified in interpreting the phrase "more consistent shape perception" as equivalent to "better shape perception." Note, by the way, that "more consistent" as distinct from "facilitating" or "interfering" does not seem to be among the choices entertained in the title heading. That dichotomy has effectively disappeared by the end of the passage, because it cannot be settled by the data. In what is effectively a case of "bait and switch," a notion of "consistent" performance has been substituted for "better" (facilitated) performance. Under the circumstances, I think a reader might be excused for forming the impression that the answer to the question posed in the section heading was that specularity facilitates. The idea that more "consistent" performance corresponds to overall worse performance is not something that one would naturally assume.

In any event, at a couple of points in a very opaque and incoherent text, the authors share with us the fact that "Surprisingly, overall [so-called] specular objects tended to be less discriminable than [so-called] matte-textured objects" (p. 10) and that "for in-depth rotations, discriminability of specular objects was overall lower than that of matte shapes, and for viewing axis rotations, it never exceeded that of mattte objects." (p. 11). (As far as I can see, these facts are not made clear in the Results section).

So, to make lemonade out of lemons, the authors treat the (unexplained) lesser variability (across conditions) but overall worse performance in the "specular" case across conditions as advantageous relative to the (unexplained) overall better performance but greater variability (across conditions) of the "matte" case. The one specific claim, post hoc, that motion parallax is not used in the specular case, but is used in the matte case, is only a somewhat bizarre and superficial attempt to explain the worse performance (but not the greater "consistency") of the specular case.

What is, in fact, the case is that the premises (tbd) and methods (tbd) of this project are sloppy, the analysis confused, forced, and misleading.

On 2017 Jun 20, Lydia Maniatis commented:

The term "motion parallax" figures prominently as an (oddly negative) explanatory concept in Dovencioglu et al's (2017) brief conclusion. Which is interesting, since we do not come across it in the title, abstract, introduction, methods or results sections of the paper.

It shows up for the first time in the discussion, in the form of a highly speculative suggestion aiming to explain the study's unpredicted and uninterpretable results, in particular why it failed to produce an effect that, throughout the paper, the authors create the impression it did produce.

I don't know if they deliberately meant to mislead, but until I got to a pair of isolated statements in the Discussion, I was under the impression that they were claiming to have shown that "specular motion" (a highly problematic term in its own right, as will be discussed in a subsequent comment) improved observers' ability to make veridical estimates of 3D shape. This is definitely not the case.

While they're careful not to say it outright in the abstract, and while the title, as is typical of vision papers, is uninformative, the prominence of the "specular" term might lead an unsuspecting reader to assume that the claim that results "provide an additional layer of evidence for the capacity of the visual system to exploit image information for shape inference" is referring to some value-added information provided by "specular" as opposed to non-specular, stimuli.

Similarly, the text of the final section of the introduction, titled "Does specular flow facilitate or interfere with 3D shape extimation? " might give the impression that the former is the case: For example:

"...we focus on...whether specular flow...can provide better information on 3D shape than optic flow...specular flows are directly related to 3D curvature and seem to be less sensitive to the particular motion of the object, whereas optic flows vary more substantially with the latter. Thus, if a perceptual task required observers to make judgments abouta an objects' 3D curvature structure...one would expect more consistent shape perception across changes in object rotation axis."

Given the lead-up, I think a reader would be justified in interpreting the phrase "more consistent shape perception" as equivalent to "better shape perception." Note, by the way, that "more consistent" as distinct from "facilitating" or "interfering" does not seem to be among the choices entertained in the title heading. That dichotomy has effectively disappeared by the end of the passage, because it cannot be settled by the data. In what is effectively a case of "bait and switch," a notion of "consistent" performance has been substituted for "better" (facilitated) performance. Under the circumstances, I think a reader might be excused for forming the impression that the answer to the question posed in the section heading was that specularity facilitates. The idea that more "consistent" performance corresponds to overall worse performance is not something that one would naturally assume.

In any event, at a couple of points in a very opaque and incoherent text, the authors share with us the fact that "Surprisingly, overall [so-called] specular objects tended to be less discriminable than [so-called] matte-textured objects" (p. 10) and that "for in-depth rotations, discriminability of specular objects was overall lower than that of matte shapes, and for viewing axis rotations, it never exceeded that of mattte objects." (p. 11). (As far as I can see, these facts are not made clear in the Results section).

So, to make lemonade out of lemons, the authors treat the (unexplained) lesser variability (across conditions) but overall worse performance in the "specular" case across conditions as advantageous relative to the (unexplained) overall better performance but greater variability (across conditions) of the "matte" case. The one specific claim, post hoc, that motion parallax is not used in the specular case, but is used in the matte case, is only a somewhat bizarre and superficial attempt to explain the worse performance (but not the greater "consistency") of the specular case.

What is, in fact, the case is that the premises (tbd) and methods (tbd) of this project are sloppy, the analysis confused, forced, and misleading.

On 2017 Jun 21, Lydia Maniatis commented:

(Comment #2)

The three sentences of the conclusion (which I annotate below, in its entirety, reflects the article's utter lack of content:

1. *"Motion information generated by moving specularities across a surface is used by human observers when judging the bumpiness of 3D shapes." *

The boundaries of specularities are effectively contour lines. It would be thoroughly unrealistic to predict that they would not play a role, moving or not. And the observation had already been made.

1. "In the presence of specular motion, observers tend to not rely on the motion parallax information generated by the matte-textured reflectance component."

The two parts of this sentence seem to be a non-sequitur - how could observers of specular motion employ information generated by matte-textured objects (i.e. objects other than the ones they were observing)? What the authors mean to say is that observers don't use the motion parallax info generated by the specular stimulus. While they frame this as though it were an actual finding, it is, as discussed above, a purely speculative attempt to explain the poorer performance with specular objects.

1. *"This study further highlights how 3D shape, surface material, and object motion interact in dynamic scenes." *

It really doesn't, given the mixed results and failed predictions. It couldn't for a number of other reasons, discussed below.

1. All of the heavy lifting in this article is done by computer programmers, whose renderings are supposed to qualify as "specular" "specular motion" "matte-textured" etc. These renderings rest on theoretical assumptions most of which are never made explicit. They are, however, inadequate; we learn that observers sometimes saw the moving specular stimuli as non-rigid. This is a problem. There is no objective description of the phenomenon "specular object in motion around an axis" other than "objects generated by this particular program." Is there any doubt that results would have been different if the renderings had accurately mimicked the physical phenomenon? If the surface of the object is seen as changing, don't this affect the "motion parallax" hypothesis? The speed with which a particular point on a surface is moving optically is confounded with the speed with which it is moving on its own.

2. The so-called matte-textured objects appeared purely reflective when not in motion. The apparent specularities were "stuck-on" so that they moved with the surface. I have never seen a matte surface with this characteristic. I would be curious to see the in-motion renderings, because I cannot imagine what they look like. What is clear is that a simple reference to matte textured objects is not appropriate. We are talking about a different phenomenon, which may not correspond to any physically actualizable one. This latter fact wouldn't matter if the theoretical framework were tight enough that such stimuli allowed isolation of some particular factor of interest. Here, however, it is just means that "matte" doesn't mean what it normally is thought to mean.

3. Observers were confused about the meaning of the term "bumpiness." Stimuli involve hills and ridges of various extents as well as varying apparent heights. The authors were interested in height. They instructed observers who asked for clarification (not the others) that they were interested in "the amplitude not the frequency." I would say a large hill or a wide ridge could qualify as more ample for people not thinking in terms of graphs with height in the ordinate. In other words, I think there is a observational confound between extent and height of the bumps.

4. In the introduction, the authors refer to previous papers which came to opposite conclusions. Presumably, this means that some relevant factors/confounds were not considered. But the authors don't attempt to analyze these conflicted citations, which thus merely function as window-dressing. They move on to their experiments, on the slightest and vaguest of pretexts, with poorly described stimuli and poorly controlled tasks.