On 2015 Aug 20, Lydia Maniatis commented:

The authors here report having “investigated how texture and stereo cues are integrated to perceive 3D slant.” The problem here (as with other literature dealing with the same topic) is with the unqualified references to “texture cues.”

2D “textures” are composed of 2D shapes, and 2D shape is dispositive when it comes to the perception of slant. If I start with a rectangle (or syncytium of rectangles) and I slant it so that it produces a trapezoidal 2D projection, then this projection will look like a slanted rectangle. If I start with a trapezoid and slant it so that it casts a rectangular 2D projection, then it will look like a fronto-parallel rectangle, not a slanted trapezoid. Perceived slant is a function of the shape of the projection.

Because shape is the key factor mediating apparent slant, and because textures are composed of shapes, saying that we will study the role of “texture” without specifying and controlling for the effects of the particular shapes of which the texture is composed is like proposing to investigate the role of “food cues” on blood pressure, without caring about what type of food we are employing as our “stimulus.” Such practice ensures the enduring confusion and ambiguity that characterizes slant studies, which often seem to contradict each other.

This study used “Voronoi” textures. Why? An earlier study (Todd et al, 2010) used textures composed of rectangles orthographically projected. Yet another study referred to by Saunders and Chen used Voronoi textures that were more regular than theirs. Do the authors believe that by using the type of structure the do, in which the cells' shape varies, and which has a particular degree of irregularity, they are controlling for the effects of shape? To continue my previous analogy, this would be like assuming that mixing many foods together will control for the effect of the individual nutrients on blood pressure.

If we know a variable matters, then we don't control for it by mixing it up its values, we control for it by controlling for it. “Food” in this analogy is obviously not an appropriate level of analysis, and the results will vary with the choice of “food.” The same goes for “texture.” What is left after we subtract shape? At the least, the authors should have provided some rationale for their choice of texture.

There is no doubt whatsoever that the choice of “texture” affects perceived slant. Erkelens (2013) used a “texture” composed of rectangles under perspective projection and found very good agreement with prediction, and an underestimation at all slants. Saunders and Chen found underestimation at low slants, and not at higher ones. They should explain why this low/high dichotomy should be found in their particular conditions, including their choice of pattern, and not in, e.g., Erkelens'. (Are they suggesting that by using “Voronoi” patterns, they have isolated the role of “texture,” independently of shape, and that their results are more valid?)

The authors' decision to deal with the potential role of the outline containing their texture by randomly varying this shape reflects the tendency to embed confounds in the data in the hope that they will average out (rather than distort or flatten the results – averaging black and white makes grey) rather than to confront and control for them head on.

Of course, they found that “texture” cues and disparity cues are somehow integrated (the claim that they are “optimally” integrated is difficult to judge, as it has been preceded by layers of speculation). The basic findings were a sure thing. The specifics of the data are uninterpretable due to lack of control of relevant variables.

On 2015 Aug 20, Lydia Maniatis commented:

The authors here report having “investigated how texture and stereo cues are integrated to perceive 3D slant.” The problem here (as with other literature dealing with the same topic) is with the unqualified references to “texture cues.”

2D “textures” are composed of 2D shapes, and 2D shape is dispositive when it comes to the perception of slant. If I start with a rectangle (or syncytium of rectangles) and I slant it so that it produces a trapezoidal 2D projection, then this projection will look like a slanted rectangle. If I start with a trapezoid and slant it so that it casts a rectangular 2D projection, then it will look like a fronto-parallel rectangle, not a slanted trapezoid. Perceived slant is a function of the shape of the projection.

Because shape is the key factor mediating apparent slant, and because textures are composed of shapes, saying that we will study the role of “texture” without specifying and controlling for the effects of the particular shapes of which the texture is composed is like proposing to investigate the role of “food cues” on blood pressure, without caring about what type of food we are employing as our “stimulus.” Such practice ensures the enduring confusion and ambiguity that characterizes slant studies, which often seem to contradict each other.

This study used “Voronoi” textures. Why? An earlier study (Todd et al, 2010) used textures composed of rectangles orthographically projected. Yet another study referred to by Saunders and Chen used Voronoi textures that were more regular than theirs. Do the authors believe that by using the type of structure the do, in which the cells' shape varies, and which has a particular degree of irregularity, they are controlling for the effects of shape? To continue my previous analogy, this would be like assuming that mixing many foods together will control for the effect of the individual nutrients on blood pressure.

If we know a variable matters, then we don't control for it by mixing it up its values, we control for it by controlling for it. “Food” in this analogy is obviously not an appropriate level of analysis, and the results will vary with the choice of “food.” The same goes for “texture.” What is left after we subtract shape? At the least, the authors should have provided some rationale for their choice of texture.

There is no doubt whatsoever that the choice of “texture” affects perceived slant. Erkelens (2013) used a “texture” composed of rectangles under perspective projection and found very good agreement with prediction, and an underestimation at all slants. Saunders and Chen found underestimation at low slants, and not at higher ones. They should explain why this low/high dichotomy should be found in their particular conditions, including their choice of pattern, and not in, e.g., Erkelens'. (Are they suggesting that by using “Voronoi” patterns, they have isolated the role of “texture,” independently of shape, and that their results are more valid?)

The authors' decision to deal with the potential role of the outline containing their texture by randomly varying this shape reflects the tendency to embed confounds in the data in the hope that they will average out (rather than distort or flatten the results – averaging black and white makes grey) rather than to confront and control for them head on.

Of course, they found that “texture” cues and disparity cues are somehow integrated (the claim that they are “optimally” integrated is difficult to judge, as it has been preceded by layers of speculation). The basic findings were a sure thing. The specifics of the data are uninterpretable due to lack of control of relevant variables.