Pith. sign in

REVIEW 4 major objections 8 minor 142 references

A Turing Test for Artificial Nets devoted to model Human Vision

T0 review · 4 major / 8 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read A compact set of ten low-level vision facts can separate artificial vision networks by how closely they mimic human early vision.

desk verdict The Decalogue is a genuinely useful low-level benchmark for vision models, but the paper's specific ranking claim rests on an unvalidated Euclidean read-out and should be treated as illustrative, not established. read the letter →

arxiv 2502.00721 v3 pith:TP3CDSVM submitted 2025-02-02 q-bio.NC

classification q-bio.NC
keywords low-levelvisionTuringtestpsychophysicsreceptivefieldsdivisivenormalizationcontrastmaskingmodelevaluationspatio-chromatic
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that claims about deep networks modeling human vision remain unproven for low-level vision, and proposes a direct check: a fixed set of ten calibrated stimulus series, each tied to a documented human visual fact, that any image-computable network can be fed. The test covers spectral sensitivities, brightness and color saturation, contrast sensitivity functions, receptive-field shapes, and nonlinear behaviors such as saturation, crispening, energy masking, frequency masking, and orientation masking. Applied to three networks of similar architecture, the test ranks a parametric model whose nonlinearities were psychophysically tuned as the most human-like, with the quality-estimation network second and the segmentation network third. The paper's methodological claim is that even this qualitative application of the test is enough to produce a meaningful ranking and to point at specific architectural corrections.

What carries the argument

The load-bearing machinery is the Decalogue: ten sets of calibrated digital stimuli that probe the retina-V1 information bottleneck, each standing for a documented human visual fact. The read-out is deliberately simple—the Euclidean departure between a model's response to a test pattern on a background and its response to the isolated background—and the comparison is made against either standardized human sensitivity surfaces or qualitative curve shapes, namely saturating responses that shift monotonically with background energy, frequency, or orientation. The nonlinear behaviors the test targets are classically produced by divisive normalization, a gain-control operation that divides each neuron's response by a pooled measure of neighboring activity, and the top-ranked model uses parametric versions of that operation tuned by the Maximum Differentiation method, a psychophysical procedure that searches for stimulus directions where model behavior differs most.

What would settle it

Re-run the same three-model comparison with alternative read-outs, for example a linear classifier probe at the same layers or a Minkowski distance with exponent other than 2, and check whether the model ordering changes; the paper's own discussion of read-out choice implies it could. A more direct test would measure human detection thresholds on the exact Decalogue stimuli and check whether the Euclidean read-out predicts those thresholds before relying on the qualitative curves.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a low-level Turing test built from ten classical retina-V1 facts separates three deep networks in a way that tracks their training histories, and that the model with psychophysically tuned divisive-normalization nonlinearities comes out closest to human behavior. The advantage is most visible in nonlinear spatio-chromatic responses: the leading model reproduces saturating contrast responses, the correct frequency ordering, and the right direction of masking by energy, frequency, and orientation, while the other two show expansive or inverted curves, wrong chromatic scaling, and masking that increases rather than decreases visibility. The paper is careful to note that even the leading model fails some feature-interaction facts, such as color induction and some texture-masking details, and it treats these failures as diagnostics for where its parameters should change. It also stresses that read-out location matters, showing that in the parametric model the spectral sensitivities are human-like at the first layer but degrade at deeper layers.

Load-bearing premise

The test's rankings depend on treating the Euclidean distance between a model's response to a test on a background and its response to the isolated background as a faithful measure of how visible the test is to a human, and the paper does not validate that read-out against human thresholds.

Editorial extensions

If this is right

  • If the test is adopted, a vision network can be screened for low-level biological plausibility without high-level task benchmarks or complicated read-out decoders.
  • The qualitative version of the test is sufficient to separate the three models, so quantitative automation is not a prerequisite for model selection.
  • For the parametric model, failures on feature-interaction facts indicate which divisive-normalization kernels to adjust, making the test diagnostic rather than merely evaluative.
  • The same stimuli can serve as fine-tuning targets after conventional training, so standard goals and human-like constraints can be alternated during optimization.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Since the paper itself grants that read-out choice is arbitrary and can change rankings, a stronger version of the test would report the model ordering under several read-outs, such as classifier probes and non-Euclidean metrics.
  • A natural quantitative extension would assign weights to the ten facts and collapse the qualitative scores into one human-alignment index that could be used as a training loss.
  • One could also measure human detection or discrimination thresholds for the exact Decalogue stimuli and test directly whether the Euclidean read-out is the right proxy, rather than taking it as given.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 8 minor

Summary. The paper proposes a low-level 'Turing test' for image-computable models of human vision: a Decalogue of ten psychophysical and physiological facts about color, contrast sensitivity, receptive fields, and nonlinear masking, together with calibrated stimuli and code to generate them. To illustrate the test, the authors apply it to three deep models: BioMultiLayer, a parameterized cascade of linear-nonlinear stages; PerceptNet, an end-to-end trained quality model with divisive normalization; and Bio U-Net, a segmentation network with the same encoder style as PerceptNet. Using a Euclidean-distance read-out of network responses and qualitative visual comparison of response curves, the paper reports that BioMultiLayer behaves most like humans, especially on nonlinear spatio-chromatic responses, and that 'even with this qualitative application ... a significant ranking is possible.' The paper also argues that the test can suggest architectural or training changes for non-human-like models.

Significance. If the proposed test were validated, it would fill a real gap: existing benchmarks such as BrainScore focus on high-level object recognition, while low-level vision facts have not been systematically compiled for image-computable model evaluation. The paper's strengths are its concrete, reproducible stimuli, the public code, and the honest framing of the evaluation as qualitative, including an explicit warning in Section 2.2 that read-out choices are arbitrary. The authors also show that the test can expose qualitative divergences among models of similar architecture, which is a useful diagnostic demonstration. However, the central quantitative claim—that the test ranks models by human alignment—rests on an unvalidated read-out and on subjective scores in Table 2; the demonstration is illustrative rather than confirmatory.

major comments (4)
  1. [Section 4.1] The read-out for visibility is the Euclidean departure between the response to test+background and the response to the isolated background, but Section 2.2 explicitly states that 'the selection of the read-out mechanism is not trivial' and that 'these (arbitrary) decisions definitely affect the characterization of the system, e.g. its frequency response.' No validation is given that Euclidean distance in the activation space of these particular networks is monotonically related to human visibility or discriminability; there is no noise model, no threshold criterion, and no calibration against human psychometric functions. Since every curve in Figures 10-12 and every entry in Table 2 depends on this single choice, the ranking of the three models is not supported unless the read-out is validated or the results are shown to be robust across a family of read-outs (e.g., different norms, Minkowski exponents, or linear classifiers).
  2. [Table 2 and Section 4.2.5] The claim that 'even with this qualitative application of the proposed test ... a significant ranking is possible' is not operationalized: the symbols in Table 2 are assigned by the authors with no defined scoring rubric, no inter-rater reliability, and no statistical test. The word 'significant' appears to mean 'clear to the authors' rather than a statistically supported statement. To make the ranking reproducible, the authors should either (a) define a quantitative distance between model response curves and reference human curves (e.g., a goodness-of-fit measure with tolerance bands), (b) report the sensitivity of the ranking to plausible variations in the qualitative judgments, or (c) explicitly restrict the claim to an illustrative demonstration rather than a ranking.
  3. [Section 4.1 and Section 4.2.5] The evaluation is partly circular for the BioMultiLayer model: its linear parts were built to contain the tested spectral sensitivities and Center-surround/Gabor receptive fields, and its nonlinearities were tuned on contrast masking and distortion data that overlap with the Decalogue (stages 1 and 4 tuned to reproduce distortion and contrast masking; stages 2 and 3 tuned via Maximum Differentiation). The paper acknowledges the linear part of this issue in Section 4.2.5, but the same concern applies to the nonlinear facts: if a model was fit to masking phenomena, observing masking in that model is not an independent success. The authors should explicitly separate facts used in model design/tuning from held-out facts and present the ranking separately for fitted versus emergent properties.
  4. [Figures 10-12 and Section 4.2] The comparative claims rest entirely on single realizations of the models, without sensitivity analysis. For the two trained networks, random initializations, training data, or choice of read-out layer can materially change CSFs and masking behavior; the paper reports no variation across seeds or layers. Even for the deterministic BioMultiLayer, the parameter values are the result of a fitting procedure, and no bootstrap or perturbation analysis is shown. Without such robustness checks, the observed differences between models cannot be distinguished from incidental implementation choices.
minor comments (8)
  1. [Section 1.1] In the prologue, 'questionare' should be 'question are' (or 'questionnaire').
  2. [Section 2.1] The sentence 'However, it is not obvious how how these tasks may be combined' contains a duplicated word 'how'.
  3. [Table 1] In the Modality column, the entries for facts 3 through 10 appear as 'Text'; this should be 'Texture'.
  4. [Section 3.2.2] The phrase 'chromati contexts' should read 'chromatic contexts'.
  5. [Section 4.1] The description of the read-out would be clearer if it stated whether the Euclidean distance is computed on raw activation vectors or on some normalized version; the units and scale of activations affect the numerical values and hence the curves in Figures 10-12.
  6. [Acknowledgments] The sentence 'The invited talk ... was funded be the University of Bristol' should say 'funded by'.
  7. [References] Reference [13] is a personal communication with no year or affiliation; please provide a proper citation or remove it.
  8. [Throughout] The model name 'Bio U-Net' appears as 'Bio-UNet' in several places; please use a single consistent spelling.

Circularity Check

2 steps flagged · score 6.0 of 10

The central ranking is partly a report of BioMultiLayer's construction and fitting data: linear facts are built in, and the masking facts were used to tune its nonlinearities.

  1. self definitional [Section 4.2.1 'Spectral sensitivities and color responses (properties 1 and 2)'; Section 4.2.5 'Summary of results']
    "In this network the first linear layer has achromatic and opponent channels defined by construction so the Vλ [84] (subplot A) and the opponent curves of Jameson & Hurvich [106] (subplot B) are trivially obtained. ... It is not surprising that the parametric model (the BioMultiLayer) has bigger alignment in the linear parts (properties 1 and 5) since sensitivities and center-surround and Gabor receptive fields were parametrically built in that model model."

    Property 1 and the receptive-field part of property 5 are not independent checks of human alignment for BioMultiLayer: the achromatic/opponent channels and the center-surround/Gabor kernels are part of the model's definition. The paper's own wording ('trivially obtained', 'parametrically built in') shows that BioMultiLayer's top scores on these facts are restatements of its construction, so those rows of Table 2 do not provide independent evidence that the Decalogue can discriminate models. The abstract's claim that this model is closer 'in terms of receptive fields' is therefore true by construction for the linear part.

  2. fitted input called prediction [Section 4.1 'Networks and experimental setting'; Section 4.2.4 'Energy masking and feature masking']
    "The nonlinearities of the 2nd and 3rd stages of the model were tuned via the psychophysical method of Maximum Differentiation in [32]. And the nonlinear parts of the 1st and 4th stages were tuned to reproduce subjective opinions on distortion and contrast masking facts [33, 34]."

    The Decalogue's nonlinear facts include the contrast-masking behavior on which BioMultiLayer's nonlinearities were fitted: fact 8 is energy masking and facts 9-10 are frequency and orientation cross-masking, while the model's stage 4 is described as 'wavelet analysis and cross-masking between textures' whose nonlinear parts were tuned to 'reproduce subjective opinions on distortion and contrast masking facts.' The paper reports fact 8 as 'qualitatively human' for BioMultiLayer and gives that model the best nonlinear scores in Table 2, so the abstract's emphasis on its 'nonlinear behavior for spatio-chromatic patterns' is in part a report of fit quality, not an independent prediction.

full rationale

The Decalogue itself is an honest compilation of classical low-level facts, and the paper is transparent about BioMultiLayer's constructed nature: it says the linear parts (spectral sensitivities, opponent channels, center-surround and Gabor receptive fields) are built in, and that the nonlinear parts were fitted to distortion and contrast-masking data. That transparency prevents a score of 8-10. However, the central claim that BioMultiLayer is closest to humans, and that the test yields a significant ranking, is partially forced: properties 1 and 5 are true by construction for BioMultiLayer, and its advantage on the nonlinear masking facts reproduces the contrast-masking data used to fit its nonlinearities. The paper does contain independently informative results, such as the CSF scaling from sinusoidal responses ('an original check done here since they were not built in') and the relative ordering of PerceptNet versus Bio-U-Net, so the circularity is partial, not total. Separately, the ranking depends on an unvalidated Euclidean read-out: Section 4.1 defines responses as the 'Euclidean departure between the response to test+background with regard to the response to the isolated background', while Section 2.2 warns that 'the selection of the read-out mechanism is not trivial' and that arbitrary decisions 'definitely affect the characterization of the system.' That is a validity limitation on the ranking rather than a construction-level circularity, but it compounds the two reduction steps above.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. The free parameters are the fitted parameters of the BioMultiLayer model and the hand-assigned qualitative scores; the evaluation depends on these as well as on the assumed read-out (Euclidean distance) and the assumed adequacy of the Decalogue as a characterization of early vision.

free parameters (3)
  • BioMultiLayer divisive normalization parameters (stages 2 and 3) = not reported here (see Martinez et al. 2018)
    Tuned via Maximum Differentiation to human contrast masking data; these values are reused in the paper's evaluation, so the model's human-like masking behavior is partly a restatement of the fit.
  • BioMultiLayer nonlinearity parameters (stages 1 and 4) = not reported here (see Martinez et al. 2018, 2019)
    Fit to reproduce subjective opinions on distortion and contrast masking facts, which overlap with Decalogue facts 6-10.
  • Qualitative similarity scores in Table 2 = symbols assigned by authors
    The paper's ranking is based on hand-assigned qualitative scores with no quantitative threshold, error bars, or independent evaluation.
assumptions (4)
  • ad hoc to paper Euclidean distance between model responses to test+background and background alone is a valid visibility read-out.
    Adopted in Section 4.1 without validation; the paper itself (Section 2.2) argues read-out choice is non-trivial and can change rankings.
  • domain assumption The selected ten facts characterize the retina-V1 adaptive information bottleneck.
    The paper claims the Decalogue describes the bottleneck; this is a domain assumption drawn from classical vision science, not proven here.
  • domain assumption Human qualitative behavior for non-standardized facts (2, 8, 9, 10) is as described by the authors.
    The human curves are inferred from classical literature and eye inspection, not measured in this paper.
  • ad hoc to paper Feeding 128x128 digital images with assumed 64 cycles/deg sampling is a valid image-computable test for these models.
    Stated in Section 4.1 without sensitivity analysis or justification for each model's native input scale.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Turing Test for Artificial Nets devoted to model Human Vision." pith.science (2026). https://pith.science/paper/TP3CDSVM

@misc{pith2026250200721,
  author       = {Pith},
  title        = {Pith review of: A Turing Test for Artificial Nets devoted to model Human Vision},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TP3CDSVM}},
  note         = {Machine review of arXiv:2502.00721}
}
read the original abstract

In our invited talk at the AI Evaluation Workshop of the University of Bristol back in June 2022 we argued that, despite claims about successful modeling of the visual brain using ANNs, the problem is far from being solved (even for low-level vision). Open issues include: where should we read from ANNs to reproduce human behavior?, this ad-hoc read-out is part of the brain model or not?, should we use artificial psychophysics or artificial physiology?, artificial experiments should literally match the experiments in humans?. There is a clear need of rigorous procedures for experimental tests for ANNs models of the visual brain, and more generally, to understand ANNs devoted to generic vision tasks. Following our experience in using low-level facts from Visual Neuroscience in Image Processing, we presented the idea of developing a low-level dataset compiling the basic spatio-temporal and chromatic facts that are known to happen in the retina-V1 pathway, and they are not currently available in existing databases such as BrainScore. In our results we checked the behavior of three recently proposed models with similar architecture: (1) A parametric model tuned via Maximum Differentiation [Malo & Simoncelli SPIE 15, Martinez et al. PLOS 18, Martinez et al. Front. Neurosci. 19], (2) A non-parametric model, the PerceptNet, tuned to maximize the correlation with human opinion on subjective distortions [Hepburn et al. IEEE ICIP 20], and (3) A model with the same encoder as PerceptNet, but tuned for segmentation (published later as Hernandez-Camara et al. Patt.Recogn.Lett. 23, Hernandez-Camara et al. Neurocomp. 25). Results on 10 compelling psycho/physio visual facts show that the first model is the one with closer behavior to the humans in terms of receptive fields, but more interestingly, on the nonlinear behavior for spatio-chromatic patterns of a range of luminances and contrasts.

Figures

Figures reproduced from arXiv: 2502.00721 by the authors.

Figure 1
Figure 1. Image Denoising, Image Compression, Image Segmentation and Image Classification architectures with (eventually) biological correlates in the LGN, the V1 and beyond. However, it is not obvious how how these tasks may be combined to explain biological vision. 3 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Given a deep model successfully trained for some visual task, the read-out location and read-out mechanism (or decoder) is important to assess its biological plausibility. 4 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. In artificial physiology (left) and in artificial psychophysics (right), the arbitrary decoder to read-out model activations is critical. more sophisticated techniques such as reverse correlation which are used both in physiology [82] and in psychophysics [83], and these are not yet widely used in machine learning. Regarding the experimental setting, should one go for a literal reproduction of the experiments with h… view at source ↗
Figures from the paper (15 more)
Figure 4
Figure 4. Figure 4: In measuring the spectral sensitivity of certain elements of a network one may try a literal reproduction of the human psychophysics (left) or an idealized experiment (right). The literal reproduction could be done through matching experiments [84]: finding the ratio o…
Figure 5
Figure 5. Figure 5: show a series of these stimuli in the (linearly spaced range of luminance [0.5, 120] cd/m2 on (linearly spaced) backgrounds of luminance in the range [1, 160] cd/m2 . These stimuli are easily generated in digital levels (i.e. ready to feed conventional artificial model…
Figure 6
Figure 6. Figure 6: shows colorimetrically calibrated stimuli with such deviations (in the range [-20, 20] of the linear RG and YB tristimulus values of the Jameson and Hurvich space) in different backgrounds, which are easy to generate and modify by using the code provided in this work6 …
Figure 7
Figure 7. Figure 7: shows Gaussian windowed test noise patches of 4 cycles/degree (cpd) in images subtending 1 degree with average luminance of 50 cd/m2 and linearly spaced RMSE contrasts (from left to right) in the range [0, 0.3]. The different rows show the same tests on different backg…
Figure 8
Figure 8. Figure 8: shows 6 cpd horizontal Gabor patches with average luminance of 50 cd/m2 and RMSE contrast increasing linearly from left to right in the range [0, 0.3]. These Gabor patches are shown on top of band-pass noise of contrast 0.2, with the same frequency, but different orien…
Figure 9
Figure 9. Figure 9: The proposed method: feed the model with series of images, compute responses (using the simplest possible read-out mechanism) and make quantitative comparisons with standard sensitivity surfaces or qualitative comparisons checking the nonlinearitiy using different adap…
Figure 10
Figure 10. Figure 10: Spectral sensitivities of the considered models (top) and corresponding responses to luminance and linear deviations from white in the cardinal red-green and yellow-blue directions (bottom). On the one hand, knowledge of the standard spectral sensitivity, the CIE Vλ c…
Figure 5
Figure 5. Figure 5: In the chromatic cases, responses are computed with tests on an achromatic background (black [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 11
Figure 11. Figure 11: Achromatic and chromatic contrast sensitivities of the considered models (top subplots A, B, C), and different sets of receptive fields computed in different ways at different depths of the models (bottom). See text for details on the psychophysical and the physiologi…
Figure 12
Figure 12. Figure 12: Contrast responses of the considered models in different masking conditions and for achromatic and chromatic textures of different frequencies. The color code (indicated in the subplots corresponding to Percepnet, but applicable to the equivalent curves of the other m…
Figure 8
Figure 8. Figure 8: For this last example, the BioMultiLayer model gets bigger attenuation for the background of the same orientation, particularly for high frequency (see the red curves), and the other orientations lead to responses that are between the no-mask condition (in blue) and th…
Figure 13
Figure 13. Figure 13: (Top) Human-like saturation behavior in contrast response happening in generic shallow autoen￾coders [21]. (Bottom) Human-like behavior obtained in image segmentation U-Nets when they are equipped with bio-inspired Divisive Normalization to improve their performance […
Figure 14
Figure 14. Figure 14: Examples of errors happening in the model [33, 34, 35]. Left: too sharp chroma nonlinearity in achromatic context -wrong gray curve- (taken from [PITH_FULL_IMAGE:figures/full_fig_p019_14.png]
Figure 15
Figure 15. Figure 15: Left: Adaptation in the CSFs in autoencoders obtained from training on the proper (colormetrically calibrated) environments for color adaptation [21]. Right: Improved contrast perception by using bio-inspired Divisive Normalization (with adaptive contrast responses su…
Figure 16
Figure 16. Figure 16: Top: Image PDF factorization from the contrast nonlinearites (div. norm.) illustrated in Figs. 7 and 8, as shown in [103]. Bottom: autoencoders trained with distortion metrics based on the contrast nonlinearities described in our proposal capture natural image statist…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

142 extracted references · 75 canonical work pages

  1. [1]

    Deep problems with neural network models of human vision

    Jeffrey S Bowers, Gaurav Malhotra, Marin Dujmovi ´c, Milton Llera Montero, Christian Tsvetkov, Valerio Biscione, Guillermo Puebla, Federico Adolfi, John E Hummel, Rachel F Heaton, et al. Deep problems with neural network models of human vision. Behavioral and Brain Sciences, 46:e385, 2023

  2. [2]

    Biscione et al

    V . Biscione et al. MindSet: Vision. a toolbox for testing DNNs on key psychological experi- ments. arXiv preprint arXiv:2404.05290, 2024

  3. [3]

    Do computer vision foundation models learn the low-level characteristics of the human visual system?

    Yancheng Cai, Fei Yin, Dounia Hammou, and Rafal Mantiuk. Do computer vision founda- tion models learn the low-level characteristics of the human visual system? CVPR ArXiV: 2502.20256, 2025

  4. [4]

    Bampis, and Rafał K

    Dounia Hammou, Yancheng Cai, Pavan Madhusudanarao, Christos G. Bampis, and Rafał K. Mantiuk. Do image and video quality metrics model low-level human vision? ArXiV: 2503.16264, 2025

  5. [5]

    Kubilius et al

    J. Kubilius et al. Brain-like object recognition with high-performing shallow recurrent anns. ICLR, Arxiv: 1909.06161, 2019

  6. [6]

    Spoerer, Emer C

    Johannes Mehrer, Courtney J. Spoerer, Emer C. Jones, Nikolaus Kriegeskorte, and Tim C. Kietzmann. An ecologically motivated image dataset for deep learning yields better models of human vision. Proc. Nat. Acad. Sci., 118(8):e2011417118, 2021

  7. [7]

    Zhuang, S

    C. Zhuang, S. Yan, A. Nayebi, M. Schrimpf, MC. Frank, JJ. DiCarlo, and DLK. Yamins. Unsupervised neural network models of the ventral visual stream. Proc. Nat. Acad. Sci. , 118(3):e2014196118, 2021

  8. [8]

    Storrs, T.C

    K.R. Storrs, T.C. Kietzmann, A. Walther, J. Mehrer, and N. Kriegeskorte. Diverse deep neural networks all predict human inferior temporal cortex well, after training and fitting. Journal of Cognitive Neuroscience, 33(10):2044–2064, 2021

Show all 142 references
  1. [9]

    Rajalingham, E.B

    R. Rajalingham, E.B. Issa, P. Bashivan, K. Kar, K. Schmidt, and J.J. DiCarlo. Large-scale, high- resolution comparison of the core visual object recognition behavior of humans, monkeys, and state-of-the-art deep artificial neural networks. Journal of Neuroscience, 38(33):7255–...

  2. [10]

    Macpherson, A

    T. Macpherson, A. Churchland, T. Sejnowski, JJ. DiCarlo, Y . Kamitani, H. Takahashi, and T. Hikida. Natural and artificial intelligence: A brief introduction to the interplay between ai and neuroscience research. Neural Networks, 144:603–613, 2021

  3. [11]

    Cadena, GH

    SA. Cadena, GH. Denfield, EY . Walker, LA. Gatys, AS. Tolias, M. Bethge, and S. Ecker. Deep convolutional models improve predictions of macaque v1 responses to natural images. PLoS Comput. Biol., 15(4):e1006897, 2019

  4. [12]

    Burg, SA

    MF. Burg, SA. Cadena, Denfield GH., EY . Walker, AS. Tolias, M. Bethge, and S. Ecker. Learning divisive normalization in primary visual cortex.PLoS Comput. Biol., 16(6):e1009028, 2021

  5. [13]

    Paninsky

    L. Paninsky. Personal communication at nyu laboratory for computational vision. 2001

  6. [14]

    Marr and T

    D. Marr and T. Poggio. From understanding computation to understanding neural circuitry. Neurosci. Res. Prog. Bull., 15:470–488, 1977

  7. [15]

    D. Marr. Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. W.H. Freeman and Co., New York, 1978

  8. [16]

    From marr’s vision to the problem of human intelligence

    Tomaso Poggio. From marr’s vision to the problem of human intelligence. MIT-CBMM Memos, (118), 09/2021 2021

  9. [17]

    Marr’s three levels of analysis are useful as a framework for neuroscience

    Máté Lengyel. Marr’s three levels of analysis are useful as a framework for neuroscience. The Journal of Physiology, 602(9):1911–1914, 2024. 22

  10. [18]

    J.W. Pillow. Cross talk opposing view: Marr’s three levels of analysis are not useful as a framework for neuroscience. The Journal of Physiology, 602(9):1915–1917, 2024

  11. [19]

    Malo and P

    J. Malo and P. Hernández-Cámara. A separate theory-on-top level may be inspiring, but it is neither separate nor enough. The Journal of Physiology., 602(9):1918–1918, 2024

  12. [20]

    Gomez-Villa, A

    A. Gomez-Villa, A. Martin, J. Vazquez, M. Bertalmío, and J. Malo. Color illusions also deceive CNNs for low-level vision tasks: Analysis and implications. Vision Research, 176:156–174, 2020

  13. [21]

    Contrast sensitivity functions in autoencoders

    Qiang Li, Alex Gomez-Villa, Marcelo Bertalmío, and Jesús Malo. Contrast sensitivity functions in autoencoders. Journal of Vision, 22(6), 2022

  14. [22]

    Dissecting the effectiveness of deep features as metric of perceptual image quality

    Pablo Hernández-Cámara, Jorge Vila-Tomás, Valero Laparra, and Jesús Malo. Dissecting the effectiveness of deep features as metric of perceptual image quality. Neural Networks, 185:107189, 2025

  15. [23]

    On the relation between statistical learning and perceptual distances

    Alexander Hepburn, Valero Laparra, Raul Santos-Rodriguez, Johannes Ballé, and Jesus Malo. On the relation between statistical learning and perceptual distances. In International Conference on Learning Representations, 2022

  16. [24]

    Malo and J

    J. Malo and J. Gutiérrez. V1 non-linear properties emerge from local-to-global non-linear ICA. Network: Computation in Neural Systems, 17(1):85–102, 2006

  17. [25]

    Laparra, S

    V . Laparra, S. Jiménez, G. Camps-Valls, and Jesús Malo. Nonlinearities and adaptation of color vision from Sequential Principal Curves Analysis. Neural Comp., 24(10):2751–2788, 2012

  18. [26]

    Laparra and J

    V . Laparra and J. Malo. Visual aftereffects and sensory nonlinearities from a single statistical framework. Frontiers in Human Neuroscience, 9:557, 2015

  19. [27]

    H.B. Barlow. Sensory mechanisms, the reduction of redundancy, and intelligence. Proc. of the Nat. Phys. Lab. Symposium on the Mechanization of Thought Process, (10):535–539, 1959

  20. [28]

    H.B. Barlow. Possible principles underlying the transformation of sensory messages. In W A Rosenblith, editor,Sensory Communication, pages 217–234. MIT Press, Cambridge, MA, 1961

  21. [29]

    H.B. Barlow. Redundancy reduction revisited. Network: Computation in Neural Systems , 12:241–253, 2001

  22. [30]

    J. Malo, J. Gutiérrez, and J. Rovira. Perturbation analysis of the changes in V1 receptive fields due to context. Gordon Research Conference: Sensory Coding and the Natural Environment, 2004

  23. [31]

    H. Barlow. Personal communication at the GRC sens. coding nat. env. 2004

  24. [32]

    Simoncelli

    Jesús Malo and Eero P. Simoncelli. Geometrical and statistical properties of vision models obtained via maximum differentiation. In Bernice E. Rogowitz, Thrasyvoulos N. Pappas, and Huib de Ridder, editors, Human Vision and Electronic Imaging XX, volume 9394 of Society of Photo...

  25. [33]

    Martinez, P

    M. Martinez, P. Cyriac, T. Batard, M. Bertalmío, and J. Malo. Derivatives and inverse of cascaded linear+nonlinear neural models. PLOS ONE, 13(10):1–49, 10 2018

  26. [34]

    Martinez, M Bertalmío, and J

    M. Martinez, M Bertalmío, and J. Malo. In praise of artifice reloaded: Caution with natural image databases in modeling vision. Front. Neurosci. doi: 10.3389/fnins.2019.00008, 2019

  27. [35]

    Malo, JJ

    J. Malo, JJ. Esteve-Taboada, and M Bertalmío. Cortical divisive normalization from wilson- cowan neural dynamics. J. Nonlinear Sci., 34(2):35, 2024

  28. [36]

    Hepburn, V

    A. Hepburn, V . Laparra, J. Malo, R. McConville, and R. Santos-Rodriguez. Perceptnet: A human visual system inspired neural network for estimating perceptual distance. In IEEE ICIP, pages 121–125, 2020. 23

  29. [37]

    Ballé, V

    J. Ballé, V . Laparra, and EP. Simoncelli. End-to-end optimized image compression. ICLR ArXiV:1611.01704, 2017

  30. [38]

    Hernández-Cámara, J

    P. Hernández-Cámara, J. Vila-Tomás, V . Laparra, and J. Malo. Neural networks with divisive normalization for image segmentation. Patt. Recogn. Lett., 173:64–71, 2023

  31. [39]

    Hernández-Cámara, J

    P. Hernández-Cámara, J. Vila-Tomás, P. Dauden-Oliver, N. Alabau-Bosque, V . Laparra, and J. Malo. Why divisive normalization works in image segmentation? Neurocomputing, 649, 2025

  32. [40]

    Foundations of Computer Vision

    Antonio Torralba, Phillip Isola, and William T Freeman. Foundations of Computer Vision. MIT Press, 2024

  33. [41]

    J. J. Atick, Z. Li, and A. N. Redlich. Understanding retinal color coding from first principles. Neural Computation, 4(4):559–572, 1992

  34. [42]

    Karklin and E

    Y . Karklin and E. Simoncelli. Efficient coding of natural images with a population of noisy linear-nonlinear neurons. In Advances in Neural Information Processing Systems, volume 24. Curran Associates, Inc., 2011

  35. [43]

    Lindsey, SA

    J. Lindsey, SA. Ocko, S. Ganguli, and S. Deny. The effects of neural resource constraints on early visual representations. Int. Conf. Learn. Repr. ICLR, 2019

  36. [44]

    Understanding Vision: Theory, Models, and Data

    Li Zhaoping. Understanding Vision: Theory, Models, and Data. Oxford University Press, 05 2014

  37. [45]

    Gregory K. Wallace. The JPEG still picture compression standard. Commun. ACM , 34(4):30–44, April 1991

  38. [46]

    Malo, A.M

    J. Malo, A.M. Pons, and J.M. Artigas. Bit allocation algorithm for codebook design in vector quantization fully based on human visual system non-linearities for suprathreshold contrasts. Electronics Letters, 31(15):1229–1231, 1995

  39. [47]

    J. Malo, F. Ferri, J. Albert, J.Soret, and J.M. Artigas. The role of perceptual contrast non- linearities in image transform coding. Image & Vision Computing, 18(3):233–246, 2000

  40. [48]

    JPEG2000 Image Compression Fundamentals, Standards and Practice

    David Taubman and Michael Marcellin. JPEG2000 Image Compression Fundamentals, Standards and Practice. Springer Publishing Company, Incorporated, 2013

  41. [49]

    J. Malo, I. Epifanio, R. Navarro, and E. Simoncelli. Non-linear image representation for efficient perceptual coding. IEEE Transactions on Image Processing, 15(1):68–80, 2006

  42. [50]

    Didier J. Le Gall. The MPEG video compression algorithm. Signal Processing: Image Communication, 4(2):129–140, 1992

  43. [51]

    J. Malo, F. Ferri, J. Gutierrez, and I. Epifanio. Importance of quantizer design compared to optimal multigrid motion estimation in video coding. Electronics Letters, 36(9):507–509, 2000

  44. [52]

    J. Malo, J. Gutierrez, I. Epifanio, and F. Ferri. Perceptually weighted optical flow for motion- based segmentation in MPEG-4 paradigm. Electronics Letters, 36(20):1693–1694, 2000

  45. [53]

    Perceptual feed-back in multigrid mo- tion estimation using an improved DCT quantization

    J.Malo, J.Gutierrez, I.Epifanio, F.Ferri, and J.M.Artigas. Perceptual feed-back in multigrid mo- tion estimation using an improved DCT quantization. IEEE Transactions on Image Processing, 10(10):1411–1427, 2001

  46. [54]

    Goodale, AD.; Milner, LS

    MA. Goodale, AD.; Milner, LS. Jakobson, and DP. Carey. A neurological dissociation between perceiving objects and grasping them. Nature, 349(6305):154–156, 1991

  47. [55]

    A. D. Milner and M. A. Goodale. Separate visual pathways for perception and action. Trends Neurosci., 15:20–25, 1992

  48. [56]

    Logothetis and DL

    NK. Logothetis and DL. Sheinberg. Visual object recognition. Ann. Rev. Neurosci., 19:577– 621, 1996. 24

  49. [57]

    Kreiman, C

    G. Kreiman, C. Koch, and I. Fried. Category specific visual responses of single neurons in the human medial temporal lobe. Nat. Neurosci., 3(9):946–953, 2000

  50. [58]

    The impact on midlevel vision of statistically optimal divisive normalization in v1

    Ruben Coen-Cagli and Odelia Schwartz. The impact on midlevel vision of statistically optimal divisive normalization in v1. Journal of Vision, 13(8):13–13, 07 2013

  51. [59]

    Michelle Miller, SueYeon Chung, and Kenneth D. Miller. Divisive feature normalization improves image recognition performance in alexnet. In Int. Conf. Learn. Repres. ICLR, 2022

  52. [60]

    Akbarinia, Y

    A. Akbarinia, Y . Morgenstern, and K.R. Gegenfurtner. Contrast sensitivity function in deep networks. Neural Networks, 164:228–244, 2023

  53. [61]

    Contrast sensitivity function of multimodal vision-language models

    P Hernández-Cámara, A Gomez-Villa, JoseManuel Jaén-Lorites, J Vila-Tomás, J Malo, and V Laparra. Contrast sensitivity function of multimodal vision-language models. In 8th Cognitive Computational Neuroscience Conference, 2025

  54. [62]

    Teo and D.J

    P.C. Teo and D.J. Heeger. Perceptual image distortion.Proceedings of the SPIE, 2179:127–141, 1994

  55. [63]

    Duda and P.E

    R.O. Duda and P.E. Hart. Pattern Classification and Scene Analysis. John Wiley & Sons, New York, 1973

  56. [64]

    Norma V . S. Graham.Visual pattern analyzers. Visual pattern analyzers. Oxford University Press, New York, NY , US, 1989. Pages: xvi, 646

  57. [65]

    Laparra, J

    V . Laparra, J. Muñoz Marí, and J. Malo. Divisive normalization image quality metric revisited. JOSA A, 27(4):852–864, 2010

  58. [66]

    Perceptual image quality assessment: From error visibility to structural similarity

    Z Wang, A C Bovik, H R Sheikh, and E P Simoncelli. Perceptual image quality assessment: From error visibility to structural similarity. IEEE Trans Image Processing, 13(4):600–612, 2004

  59. [67]

    Simoncelli

    Keyan Ding, Kede Ma, Shiqi Wang, and Eero P. Simoncelli. Image quality assessment: Unifying structure and texture similarity. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(5):2567–2581, 2022

  60. [68]

    Sheikh, A.C

    H.R. Sheikh, A.C. Bovik, and G. de Veciana. An information fidelity criterion for image quality assessment using natural scene statistics. IEEE Transactions on Image Processing, 14(12):2117–2128, 2005

  61. [69]

    Sheikh and A.C

    H.R. Sheikh and A.C. Bovik. Image information and visual quality. IEEE Transactions on Image Processing, 15(2):430–444, 2006

  62. [70]

    J. Malo. Spatio-chromatic information available from different neural layers via gaussianiza- tion. The Journal of Mathematical Neuroscience, 10(18), 2020

  63. [71]

    J. Malo, B. Kheravdar, and Q. Li. Visual information fidelity with better vision models and better mutual information estimates. Journal of Vision, 21(9):2351, 2021

  64. [72]

    Functional connectivity via total correlation: Analytical results in visual areas

    Qiang Li, Greg Ver Steeg, and Jesus Malo. Functional connectivity via total correlation: Analytical results in visual areas. Neurocomputing, 571:127143, 2024

  65. [73]

    Gatys, Alexander S

    Leon A. Gatys, Alexander S. Ecker, and Matthias Bethge. Image style transfer using convolu- tional neural networks. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2414–2423, 2016

  66. [74]

    Laparra, JE

    V . Laparra, JE. Johnson, G. Camps-Valls, R. Santos-Rodriguez, and J. Malo. Estimating Information Theoretic Measures via Multidimensional Gaussianization. IEEE Trans. Patt. Anal. & Mach. Intell., 47(02):1293–1308, 2025

  67. [75]

    Malo, JJ

    J. Malo, JJ. Esteve-Taboada, G. Aguilar, M. Maertens, and FA. Wichmann. Estimating the contribution of early and late noise in vision from psychophysical data. J. Vision, 25(1):12–12, 2025. 25

  68. [76]

    Mahendran and A

    A. Mahendran and A. Vedaldi. Visualizing deep convolutional neural networks using natural pre-images. Int. J. Comput. Vis., 120:233–255, 2016

  69. [77]

    W. Luo, Y . Li, R. Urtasun, and R. Zemel. Understanding the effective receptive field in deep convolutional neural networks. In Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc., 2016

  70. [78]

    Receptive fields of single neurones in the cat’s striate cortex

    David H Hubel, Torsten N Wiesel, et al. Receptive fields of single neurones in the cat’s striate cortex. J physiol, 148(3):574–591, 1959

  71. [79]

    Integrative action in the cat’s lateral geniculate body

    David H Hubel and Torsten N Wiesel. Integrative action in the cat’s lateral geniculate body. The Journal of physiology, 155(2):385, 1961

  72. [80]

    D.L. Ringach. Spatial structure and symmetry of simple-cell receptive fields in macaque primary visual cortex. J. Neurophysiol., 88(1):455-463, 2002

  73. [81]

    Tailby, SG

    C. Tailby, SG. Solomon, NT. Dhruv, and P. Lennie. Habituation reveals fundamental chromatic mechanisms in striate cortex of macaque. J. Neurosci., 28(5):1131–1139, 2008

  74. [82]

    Ringach and R

    D. Ringach and R. Shapley. Reverse correlation in neurophysiology. Cognit. Sci., 28(2):147– 166, 2004. Rendering the Use of Visual Information from Spiking Neurons to Recognition

  75. [83]

    Eckstein and AJ

    MP. Eckstein and AJ. Ahumada. Classification images: A tool to analyze visual strategies. J. Vision, 2(1), 2002

  76. [84]

    G Wyszecki and WS. Stiles. Color Science: Concepts and Methods, Quantitative Data and Formulae. John Wiley & Sons, New Jersey, 2000

  77. [85]

    Otazu, CA

    X. Otazu, CA. Parraga, and M. Vanrell. Toward a unified chromatic induction model.J. Vision, 10(12):5–5, 10 2010

  78. [86]

    Ware and WB

    C. Ware and WB. Cowan. Changes in perceived color due to chromatic interactions. Vision Research, 22(11):1353–1362, 1982

  79. [87]

    Gomez-Villa, A

    A. Gomez-Villa, A. Martín, J. Vazquez-Corral, M. Bertalmío, and J. Malo. Color illusions also deceive CNNs for low-level vision tasks: Analysis and implications. Vision Research, 176:156–174, November 2020

  80. [88]

    Gomez-Villa, K

    A. Gomez-Villa, K. Wang, CA. Parraga, B. Twardowski, J. Malo, J. Vazquez-Corral, and J. van de Weijer. The art of deception: Color visual illusions and diffusion models. IEEE Comp. Vis. Patt. Recogn. (CVPR), 2025

  81. [89]

    Malo and J

    J. Malo and J. Bowers. The low-level mindset: compelling low-level visual psychophysics to evaluate image computable vision models. Invited talk, Psychol. Dept. University of Bristol, 2024

  82. [90]

    Rust and JA

    NC. Rust and JA. Movshon. In praise of artifice. Nature Neurosci., 8(12):1647–1650, 2005

  83. [91]

    Schütt and FA

    HH. Schütt and FA. Wichmann. An image-computable psychophysical spatial vision model. J. Vision, 17(12):12–12, 10 2017

  84. [92]

    Bertalmío, A

    M. Bertalmío, A. Gomez-Villa, A. Martín, J. Vazquez, D. Kane, and J. Malo. Evidence for the intrinsically nonlinear nature of receptive fields in vision. Scientific Reports, 10:16277, 2020

  85. [93]

    Bertalmío, A

    M. Bertalmío, A. Durán-Vizcaíno, J. Malo, and FA. Wichmann. Plaid masking explained with input-dependent dendritic nonlinearities. Sci. Rep., 14:24856, 2024

  86. [94]

    Carney, SA

    T. Carney, SA. Klein, CW. Tyler, AD. Silverstein, B. Beutter, D. Levi, AB. Watson, AJ. Reeves, AM. Norcia, C. Chen, W. Makous, and MP. Eckstein. Development of an image/threshold database for designing and testing human vision models. In Human Vision and Electronic Imaging IV,...

  87. [95]

    Majaj, Rishi Rajalingham, Elias B

    Martin Schrimpf, Jonas Kubilius, Ha Hong, Najib J. Majaj, Rishi Rajalingham, Elias B. Issa, Kohitij Kar, Pouya Bashivan, Jonathan Prescott-Roy, Franziska Geiger, Kailyn Schmidt, Daniel L. K. Yamins, and James J. DiCarlo. Brain-score: Which artificial neural network for object ...

  88. [96]

    S. Daly. Visible differences predictor: An algorithm for the assessment of image fidelity. In A.B. Watson, editor, Digital Images and Human Vision, pages 179–206, Massachusetts, 1993. MIT Press

  89. [97]

    Watson et al

    A.B. Watson et al. Digital Images and Human Vision. MIT Press, Massachusetts, 1993

  90. [98]

    Malo, A.M

    J. Malo, A.M. Pons, and J.M. Artigas. Subjective image fidelity metric based on bit allocation of the human visual system in the dct domain. Image and Vision Computing, 15(7):535–548, 1997

  91. [99]

    A. B. Watson and J. Malo. Video quality measures based on the standard spatial observer. In IEEE Proc. Int. Conf. Im. Proc., volume 3, pages III–III, 2002

  92. [100]

    Hernández-Cámara, P

    P. Hernández-Cámara, P. Daudén-Oliver, V . Laparra, and J. Malo. Alignment of color dis- crimination in humans and image segmentation networks. Front. Psychol., V olume 15 - 2024, 2024

  93. [101]

    Olshausen and D

    B. Olshausen and D. Field. Emergence of simple-cell receptive field properties by learning a sparse code for natural images. Nature, 281:607–609, 1996

  94. [102]

    Schwartz and E.P

    O. Schwartz and E.P. Simoncelli. Natural signal statistics and sensory gain control. Nat. Neurosci., 4(8):819–825, 2001

  95. [103]

    Malo and V

    J. Malo and V . Laparra. Psychophysically tuned divisive normalization approximately factor- izes the pdf of natural images. Neural computation, 22(12):3179–3206, 2010

  96. [104]

    Visual information flow in Wilson–Cowan networks

    Alexander Gomez-Villa, Marcelo Bertalmío, and Jesus Malo. Visual information flow in Wilson–Cowan networks. Journal of Neurophysiology, 123(6):2249–2268, 2020

  97. [105]

    J. Malo. Information flow in biological networks for color vision. Entropy, 24:1442, 2022

  98. [106]

    Hurvich and Dorothea Jameson

    Leo M. Hurvich and Dorothea Jameson. An opponent-process theory of color vision. Psycho- logical Review, 64, Part 1 6:384–404, 1957

  99. [107]

    Campbell and J.G

    F.W. Campbell and J.G. Robson. Application of Fourier analysis to the visibility of gratings. Journal of Physiology, 197:551–566, 1968

  100. [108]

    K. T. Mullen. The CSF of human colour vision to red-green and yellow-blue chromatic gratings. J. Physiol., 359:381–400, 1985

  101. [109]

    Georgeson and GD

    MA. Georgeson and GD. Sullivan. Contrast constancy: deblurring in human vision by spatial frequency channels. J. Physiol., 252(3):627–656, 1975

  102. [110]

    S. Daly. Application of a noise-adaptive Contrast Sensitivity Function to image data compres- sion. Optical Engineering, 29(8):977–987, 1990

  103. [111]

    D. H. Kelly. Motion and vision. ii. stabilized spatio-temporal threshold surface. J. Opt. Soc. Am., 69(10):1340–1349, Oct 1979

  104. [112]

    Díez-Ajenjo, P

    MA. Díez-Ajenjo, P. Capilla, and MJ. Luque. Red-green vs. blue-yellow spatio-temporal contrast sensitivity across the visual field. J. Mod. Opt., 58(19-20):1736–1748, 2011

  105. [113]

    Fairchild

    M.D. Fairchild. Color Appearance Models. The Wiley-IS&T Series in Imaging Science and Technology. Wiley, 2013

  106. [114]

    crispening effect

    Paul Whittle. Brightness, discriminability and the “crispening effect”. Vision Research, 32(8):1493–1507, 1992

  107. [115]

    Laughlin

    S. Laughlin. A simple coding procedure enhances a neuron’s information capacity. Zeitschrift Für Naturforschung C, 36(9):910–912, 1981. 27

  108. [116]

    Malo and M.J

    J. Malo and M.J. Luque. ColorLab: A Matlab Toolbox for Color Science and Calibrated Color Image Processing. Univ. Valencia. https://isp.uv.es/code/vision_and_color/colorlab/content/, 2002

  109. [117]

    Vila-Tomás, P

    J. Vila-Tomás, P. Hernández-Cámara, and J. Malo. Artificial psychophysics questions classical hue cancellation experiments. Frontiers in Neuroscience, 17, 2023

  110. [118]

    Krauskopf and K

    J. Krauskopf and K. Gegenfurtner. Color discrimination and adaptation. Vision Research, 32(11):2165–2175, 1992

  111. [119]

    García, Luis Jiménez del Barco, and E

    Javier Romero, José A. García, Luis Jiménez del Barco, and E. Hita. Evaluation of color- discrimination ellipsoids in two-color spaces. J. Opt. Soc. Am. A, 10(5):827–837, May 1993

  112. [120]

    G.E Legge and J.M. Foley. Contrast masking in human vision. Journal of the Optical Society of America, 70:1458–1471, 1980

  113. [121]

    A power law for contrast discrimination

    G.E Legge. A power law for contrast discrimination. Vision Research, 18:68–91, 1981

  114. [122]

    John M. Foley. Human luminance pattern-vision mechanisms: masking experiments require a new model. J. Opt. Soc. Am. A, 11(6):1710–1719, Jun 1994

  115. [123]

    Watson and Joshua A

    Andrew B. Watson and Joshua A. Solomon. Model of visual contrast gain control and pattern masking. J. Opt. Soc. Am. A, 14(9):2379–2391, Sep 1997

  116. [124]

    Malo and J

    J. Malo and J. Gutierrez. VistaLab: The Matlab toolbox for linear spatio-temporal Vision Models. Univ. Valencia. https://isp.uv.es/code/vision_and_color/colorlab/vistalab/, 2002

  117. [125]

    Speed, and Fergus William Campbell

    John Ross, Harriet D. Speed, and Fergus William Campbell. Contrast adaptation and contrast masking in human vision. Proceedings of the Royal Society of London. Series B: Biological Sciences, 246(1315):61–70, 1991

  118. [126]

    Robert Shapley and Michael J. Hawken. Color in the cortex: single- and double-opponent cells. Vision Research, 51(7):701–717, 2011. Vision Research 50th Anniversary Issue: Part 1

  119. [127]

    Blakemore and F

    C. Blakemore and F. Campbell. On the existence of neurons selectivity sensitive to the orientation and size of retinal images. J. Physiol., 203:237–260, 1969

  120. [128]

    Martinez, LM

    M. Martinez, LM. Martinez, and J. Malo. Topographic independent component analysis reveals random scrambling of orientation in visual space. PLoS ONE, 12(6):e0178345, 2017

  121. [129]

    P. N. Loxley. The two-dimensional gabor function adapted to natural image statistics: A model of simple-cell receptive fields and sparse structure in images. Neural Computation, 29(10):2769–2799, 2017

  122. [130]

    M. U. Gutmann, V . Laparra, A. Hyvärinen, and J. Malo. Spatio-chromatic adaptation via higher-order canonical correlation analysis of natural images. PloS ONE, 9(2):e86481, 2014

  123. [131]

    Carandini and D

    M. Carandini and D. Heeger. Summation and division by neurons in visual cortex. Science, 264(5163):1333–6, 1994

  124. [132]

    Matteo Carandini and David J. Heeger. Normalization as a canonical neural computation. Nature Reviews Neuroscience, 13(1):51–62, January 2012. Number: 1 Publisher: Nature Publishing Group

  125. [133]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In F. Pereira, C.J. Burges, L. Bottou, and K.Q. Weinberger, editors, Advances in Neural Information Processing Systems, volume 25. Curran Associates, Inc., 2012

  126. [134]

    Simoncelli

    Valero Laparra, Alexander Berardino, Johannes Ballé, and Eero P. Simoncelli. Perceptually optimized image rendering. J. Opt. Soc. Am. A, 34(9):1511–1525, Sep 2017

  127. [135]

    Parametric perceptnet: A bio-inspired deep-net trained for image quality assessment

    Jorge Vila-Tomás, Pablo Hernández-Cámara, Valero Laparra, and Jesús Malo. Parametric perceptnet: A bio-inspired deep-net trained for image quality assessment. ArXiV, page 2412.03210, 2025. 28

  128. [136]

    Simoncelli and E.H

    E.P. Simoncelli and E.H. Adelson. Subband Image Coding, chapter Subband Transforms, pages 143–192. Kluwer Academic Publishers, Norwell, MA, 1990

  129. [137]

    D. Cai, GC. DeAngelis, and RD. Freeman. Spatiotemporal receptive field organization in the lateral geniculate nucleus of cats and kittens. J. Neurophysiol., 78(2):1045–1061, 1997

  130. [138]

    A.B. Watson. Detection and recognition of simple spatial forms. In O.J. Braddick and A.C. Sleigh, editors, Physical and Biological Processing of Images, volume 11 of Springer Series on Information Sciences, pages 100–114, Berlin, 1983. Springer Verlag

  131. [139]

    Martinez-Uriegas

    E. Martinez-Uriegas. Color detection and color contrast discrimination thresholds. In Proc. OSA Meeting, page 81, 1997

  132. [140]

    Gutiérrez, F

    J. Gutiérrez, F. Ferri, and J. Malo. Regularization operators for natural images based on nonlinear perception models. IEEE Tr. Im. Proc., 15(1):189–200, 2006

  133. [141]

    Hyvarinen, J

    A. Hyvarinen, J. Hurri, and PO. Hoyer. Natural Image Statistics: A Probabilistic Approach to Early Computational Vision. Springer, 2009

  134. [142]

    Watson and C.V

    A.B. Watson and C.V . Ramirez. A Standard Observer for Spatial Vision. Investig. Opht. and Vis. Sci., 41(4):S713, 2000. 29

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.