On 2016 Nov 22, Lydia Maniatis commented:

his article exhibits at least four common pathologies of the vision science literature.

First, the authors have adopted what, after Graham, I refer to as the “transparent brain hypothesis.”

Graham (1992), notes that, at a time when neuroscientists thought V1 was pretty much all there was to the visual cortex, many psychophysical experimental results perfectly matched the early descriptions of V1 neural characteristics.

Unfortunately, later evidence showed not only that there were many other hierarchically later levels of processing (V2, etc) but that even the descriptions of V1 receptive fields were overly simplistic.

How then, Graham asks, can we explain the mountains of lovely psychophysical results? Her reply is that, under certain conditions, the brain becomes “transparent” and experience directly reflects V1 activity. Teller (1984) had already described this attitude as the “nothing mucks it up proviso,” which she didn’t think was sound. Here, Kwon et al seem to believe that the use of short line segments in their stimuli causes them to directly tap into the V1 layer. (Discussions of both Teller (1984) and Graham (1992) can be found on PubPeer).

Proponents of this frankly bizarre view need, at the least, to meet the burden of explaining what happens at all the other levels of visual processing, with respect to their phenomenon of interest - bearing in mind that the same V1 receptive field activity that mediates observers' experience of their stimuli underlies, and thus is consistent with, and thus must explain, every aspect of their visual experience.

Second, and relatedly, the authors treat perception as a detection problem, rather than as an indirect, inferential process. They say: “Before providing details of our model, we will summarize some relevant characteristics of the topographical map of area V1, which make it uniquely suited for detecting closed curves on a retinal image.”

The reference to closed curves on a retinal image is unfortunate, since the stimulation of the retina is point stimulation, and curves are not a property of the proximal stimulus, but of the percept. Figure-ground segmentation is not, as I'm sure the authors are well aware, achieved by directly reading off retinal activity, in a process of detection. Any description of “contour detection” that can be developed with respect to the known properties of V1 receptive fields will be too simplistic; as a result, it will be trivial to construct any number of falsifying cases requiring a much broader - both geometrically and theoretically - perspective.

Actually, we don’t even have to try to find a falsifying case, because the authors do it for us: “But how about extracting overlapping and intersecting curves, for example, two elongated ellipses intersecting at 4 points? One of the anonymous Reviewers raised this question. The model, in its present form, does not guarantee that such [overlapping] individual ellipses can be extracted: the ambiguities at X intersections will usually not be resolved correctly…” The literature is filled with ad hoc models of this type, i.e. models that “in their present form” are inadequate to their stated purpose.

In the article highlights, the authors call theirs “the first principled explanation of the center of gravity tendency;” but one could argue that it isn’t very principled to describe a failed hypothesis as having explained anything. Perhaps the model can be fixed so as to handle a couple of overlapping ellipses, but it will almost certainly fail again soon thereafter. Sequential ad hoc fixes probably won’t result in an adequately limber model.

Finally, as is often the case with psychophysical experiments, the number of observers is very small (three), one of whom is an author. We’re told, for good measure, that one of the observers was naïve to the purpose of the experiment. Is this naivete important? If so, then what about the other two observers? If not, then why mention it?

I think what the authors refer to as "contour fragments" can also be considered "contours," i.e. they're homogeneously colored figures with a (closed) contour. That is, the "contour fragment" / "contour" dichotomy is a false one.

On 2016 Nov 22, Lydia Maniatis commented:

his article exhibits at least four common pathologies of the vision science literature.

First, the authors have adopted what, after Graham, I refer to as the “transparent brain hypothesis.”

Graham (1992), notes that, at a time when neuroscientists thought V1 was pretty much all there was to the visual cortex, many psychophysical experimental results perfectly matched the early descriptions of V1 neural characteristics.

Unfortunately, later evidence showed not only that there were many other hierarchically later levels of processing (V2, etc) but that even the descriptions of V1 receptive fields were overly simplistic.

How then, Graham asks, can we explain the mountains of lovely psychophysical results? Her reply is that, under certain conditions, the brain becomes “transparent” and experience directly reflects V1 activity. Teller (1984) had already described this attitude as the “nothing mucks it up proviso,” which she didn’t think was sound. Here, Kwon et al seem to believe that the use of short line segments in their stimuli causes them to directly tap into the V1 layer. (Discussions of both Teller (1984) and Graham (1992) can be found on PubPeer).

Proponents of this frankly bizarre view need, at the least, to meet the burden of explaining what happens at all the other levels of visual processing, with respect to their phenomenon of interest - bearing in mind that the same V1 receptive field activity that mediates observers' experience of their stimuli underlies, and thus is consistent with, and thus must explain, every aspect of their visual experience.

Second, and relatedly, the authors treat perception as a detection problem, rather than as an indirect, inferential process. They say: “Before providing details of our model, we will summarize some relevant characteristics of the topographical map of area V1, which make it uniquely suited for detecting closed curves on a retinal image.”

The reference to closed curves on a retinal image is unfortunate, since the stimulation of the retina is point stimulation, and curves are not a property of the proximal stimulus, but of the percept. Figure-ground segmentation is not, as I'm sure the authors are well aware, achieved by directly reading off retinal activity, in a process of detection. Any description of “contour detection” that can be developed with respect to the known properties of V1 receptive fields will be too simplistic; as a result, it will be trivial to construct any number of falsifying cases requiring a much broader - both geometrically and theoretically - perspective.

Actually, we don’t even have to try to find a falsifying case, because the authors do it for us: “But how about extracting overlapping and intersecting curves, for example, two elongated ellipses intersecting at 4 points? One of the anonymous Reviewers raised this question. The model, in its present form, does not guarantee that such [overlapping] individual ellipses can be extracted: the ambiguities at X intersections will usually not be resolved correctly…” The literature is filled with ad hoc models of this type, i.e. models that “in their present form” are inadequate to their stated purpose.

In the article highlights, the authors call theirs “the first principled explanation of the center of gravity tendency;” but one could argue that it isn’t very principled to describe a failed hypothesis as having explained anything. Perhaps the model can be fixed so as to handle a couple of overlapping ellipses, but it will almost certainly fail again soon thereafter. Sequential ad hoc fixes probably won’t result in an adequately limber model.

Finally, as is often the case with psychophysical experiments, the number of observers is very small (three), one of whom is an author. We’re told, for good measure, that one of the observers was naïve to the purpose of the experiment. Is this naivete important? If so, then what about the other two observers? If not, then why mention it?

I think what the authors refer to as "contour fragments" can also be considered "contours," i.e. they're homogeneously colored figures with a (closed) contour. That is, the "contour fragment" / "contour" dichotomy is a false one.