Skip to PREreview

PREreview of Colour vision is aligned with natural scene statistics at 4 months of age

by Daniel Garside
Published
DOI
10.5281/zenodo.7976117
License
CC BY 4.0

My review is a response to the V1 version on biorxiv: https://www.biorxiv.org/content/10.1101/2022.06.06.494927v1?versioned=true

Summary

  • Overall I think this is a fascinating and valuable paper. Useful data is presented, alongside a novel method for collecting such data. There is also interesting discussion of analysis of psychophysical data, pertaining to how to interpret psychometric curves (Is it reasonable to assume that attention affects only affect the plateau? Probably not)

  • This data furthers our understanding of how and why biases in colorspace exist, by showing that infants show the same bias in discrimination thresholds that have been found in adults - higher discrimination thresholds are shown for the daylight locus (roughly blue/yellow) than for the anti-daylight locus (roughly red/green).

  • My greatest criticism of the paper is that the first (key?) point of the discussion is hard to follow - it is not entirely clear to me how the statement "*The strong dependence of axis ratio on alpha [...] implies that the psychometric function describing performance as a function of saturation has both a higher mean and steeper slope for blue and yellow targets than for red and green targets*" - my understanding is that the presence of a significant effect does not necessarily imply anything specific about the shape of the function? Is this instead then a qualitative interpretation of F2A? It is quite possible that I am misunderstanding this. Either way - my suggestion is that a visual schematic might make this section easier for the reader to comprehend how this link is made.

  • Additionally, I would be curious to see a comparison between results using the local linear ‘model-free’ algorithm and a conventional Gaussian or Weibull CDF. My intuition (note: I have only skimmed the paper on the local linear method) is that these models would allow the authors to compute confidence intervals on the curves, and also meaningful parameters, all of which would be useful.

  • The extension that I would love to see (not in this paper, but in the future) would be data collected for a higher number of axes - showing that a discrimination ellipse is more extended in the top left and bottom right quadrants of MB space does not firmly convince me that it is because of the daylight locus; for that I would want to see much more precise alignment.

Minor points:

  • > and attributes of colour (Long et al., 2006) in natural scenes and within the lifetime of the individual (Schefrin & Werner, 1990)

  • F1A and 1C should have the same axes (and axis labels)

  • F1D/E/F broken out into a new figure? (split background and methodology for clarity?)

  • F2A is small/blurry - difficult to read text

  • F2A might benefit from horizontal lines at the chosen alpha values

  • F2B It might be worth explicitly noting in the legend the different axis limits for the righthand subplot, and pointing to where in the text this is discussed

  • F2A needs confidence intervals (I suspect that the uncertainty could then be inherited into the elliptical fitting also, which might avoid the use of bootstrapping...)

  • Why [0,**48**,90,...] rather than [0, 45, 90,...]? (I found later that it was because they are the daylight and anti-daylight, rather than the midpoints of the axes - I think this would benefit from clarification, either in the legend for F1 or F2)

  • F2B/C/D The different axis limits between the infants and the adults is quite confusing. Could this perhaps be plotted with a different background color (or some other visual difference) to alert the reader that there is a distinction?

  • F2D might benefit from horizontal dashed lines at axis ratio = 1

  • > axis ratio is large at low alphas and then decreases as a function of alpha

    • ...which would be expected as at alpha = 1 everything is above threshold, which would lead to an axis ratio of 1 by definition? Is this right? (Of course, only the adult group actually reaches alpha = 1)

    • - If this is so, why does F2D not plateau at 1?

  • > but maintain an axis ratio larger than one at all alphas

    • why?

  • > using a Skillings-Mack test

    • Looking at F2D for the adults (esp. considering the confidence intervals for this group) it seems surprising to me that one could confidently conclude a confident effect or alpha

  • What is the equation for the psychometric function used?

  • > has [...] a higher mean

    • in what sense?

  • > Discrimination thresholds measured using preferential looking for infants of different ages (Knoblauch et al., 2001) but owing to the time-consuming nature of preferential looking, are available for only a limited number of colour axes (see SI Development of Colour Sensitivity).

    • word missing somewhere?

  • > optimally spectrally positioned to extract colour information from the natural world.

    • - reference? Not because I don't believe it, but because I was struggling to find one recently!

  • > two axes were a reflection of this line through the vertical axis of the MacLeod-Boynton chromaticity diagram

    • Do you mean that one was the line and one was a reflection of the line through the vertical axis?

  • Typo in FM1? Fixation for 116ms? Maybe should be 160ms?

  • In FSI1 - is the minimum bound of the y-axis 0, or 7.65? If 0, why does the function not bottom out at 7.65%?

  • > If the factors determining thresholds at are identical at each alpha

    • typo

  • > such as attention and visual sensitivity vary with age.

    • comma after visual sensitivity?

  • FSI5 is interesting. Consider bumping to discussion section in main paper?

  • >but whether or not scene statistics are extracted and represented colour is currently unknown.

    • typo?

  • > is optimised in many ways

    • formatting issue?

  • > Process to calibrate perception

    • - typo?

  • > colour mechanisms(Bosten et al., 2015).

    • formatting issue? lacking space between text and reference

  • > the slope(Arazi et al., 2017

    • formatting issue? lacking space between text and reference

  • > upper asymptote(Swanson

    • formatting issue? lacking space between text and reference

Competing interests

My conflict of interests: I have closely followed the work of this group, and have a great deal of respect for their collected body of work. It is quite plausible, and definitely desirable to me, that I might collaborate with this group in the future. With this said, I shall try my best to give unbiased feedback.