CoChat

Superior colliculus neurons encode a visual saliency map during free viewing of natural dynamic video

Open access
White, Brian J., Berg, David J., Kan, Janis Y., Marino, Robert A., Itti, Laurent, Munoz, Douglas P. · 2017 · Nature Communications

The DOI matches its Crossref record; 56 of the 56 references it cites were found in a registry or an index.

Abstract

Models of visual attention postulate the existence of a saliency map whose function is to guide attention and gaze to the most conspicuous regions in a visual scene. Although cortical representations of saliency have been reported, there is mounting evidence for a subcortical saliency mechanism, which pre-dates the evolution of neocortex. Here, we conduct a strong test of the saliency hypothesis by comparing the output of a well-established computational saliency model with the activation of neurons in the primate superior colliculus (SC), a midbrain structure associated with attention and gaze, while monkeys watched video of natural scenes. We find that the activity of SC superficial visual-layer neurons (SCs), specifically, is well-predicted by the model. This saliency representation is unlikely to be inherited from fronto-parietal cortices, which do not project to SCs, but may be computed in SCs and relayed to other areas via tectothalamic pathways.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/ncomms14263

Cites

56 of 56 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Saliency and priority coding in the superior colliculus (SC). (a) Conceptual framework of saliency model. Visual input is decomposed into several topographic feature maps for luminance contrast, colour opponency, oriented edges, flicker and motion. Spatial centre-surround competition for representation in each feature map highlights locations which stand out from their neighbours. All features are integrated into a single saliency map which encodes salience in a feature- and behaviour-agnostic manner and which, combined with top–down signals, gives rise to a priority signal that controls orienting behaviour. Abbreviations: Lum: luminance; R–G: red–green colour opponency; B–Y: blue–yellow colour opponency. (b) Simplified schematic of the dominant inputs and outputs of the primate SC. SCs: superior colliculus superficial layers; SCi: superior colliculus intermediate layers.
Figure 1 Saliency and priority coding in the superior colliculus (SC). (a) Conceptual framework of saliency model. Visual input is decomposed into several topographic feature maps for luminance contrast, colour opponency, oriented edges, flicker and motion. Spatial centre-surround competition for representation in each feature map highlights locations which stand out from their neighbours. All features are integrated into a single saliency map which encodes salience in a feature- and behaviour-agnostic manner and which, combined with top–down signals, gives rise to a priority signal that controls orienting behaviour. Abbreviations: Lum: luminance; R–G: red–green colour opponency; B–Y: blue–yellow colour opponency. (b) Simplified schematic of the dominant inputs and outputs of the primate SC. SCs: superior colliculus superficial layers; SCi: superior colliculus intermediate layers.CC BY · PubMed Central
Figure 2 Saliency coding in the SC during free viewing of natural dynamic scenes. (a) Single frame of an HD clip (crosshair: eye position; annulus: receptive field (RF) of the neuron). (b) Transformation into log-polar SC space based on24. (c) Model-predicted pattern of activation across the SC map. The black regions in b represent the viewing area that extended beyond the monitor, and was blackened using non-reflective cloth (see Methods). The annulus in b and c represents the approximate point image corresponding to the RF in a. (d) Spearman correlation between model-predicted saliency and firing rate of a single SCs neuron. (e,f) Distributions of r-values for the correlation between model-predicted saliency and firing rate for the sample of 34 SCs neurons (e), and 26 SCi neurons (f). (g,h) Average firing rate (±1 standard error of the mean; s.e.m.) of two example SCs neurons as a function of time from fixation onset, when the saliency values in the RF were divided into tertiles (low, medium, high). Only fixations with duration >200 ms were included. (i) Average normalized firing rate of the 34 SCs neurons during the saccade-free epoch (0–200 ms post fixation) illustrated by the grey shaded region in g. (j,k) Average firing rate (±1 s.e.m.) of two example SCi neurons as a function of time from fixation onset, for the three saliency levels. (l) Average normalized firing rate of the 26 SCi neurons during the saccade-free epoch (0–200 ms post-fixation). Error bars in i and l indicate ±1 s.e.m. *P<0.05, paired t-test, one-tailed.
Figure 2 Saliency coding in the SC during free viewing of natural dynamic scenes. (a) Single frame of an HD clip (crosshair: eye position; annulus: receptive field (RF) of the neuron). (b) Transformation into log-polar SC space based on24. (c) Model-predicted pattern of activation across the SC map. The black regions in b represent the viewing area that extended beyond the monitor, and was blackened using non-reflective cloth (see Methods). The annulus in b and c represents the approximate point image corresponding to the RF in a. (d) Spearman correlation between model-predicted saliency and firing rate of a single SCs neuron. (e,f) Distributions of r-values for the correlation between model-predicted saliency and firing rate for the sample of 34 SCs neurons (e), and 26 SCi neurons (f). (g,h) Average firing rate (±1 standard error of the mean; s.e.m.) of two example SCs neurons as a function of time from fixation onset, when the saliency values in the RF were divided into tertiles (low, medium, high). Only fixations with duration >200 ms were included. (i) Average normalized firing rate of the 34 SCs neurons during the saccade-free epoch (0–200 ms post fixation) illustrated by the grey shaded region in g. (j,k) Average firing rate (±1 s.e.m.) of two example SCi neurons as a function of time from fixation onset, for the three saliency levels. (l) Average normalized firing rate of the 26 SCi neurons during the saccade-free epoch (0–200 ms post-fixation). Error bars in i and l indicate ±1 s.e.m. *P<0.05, paired t-test, one-tailed.CC BY · PubMed Central
Figure 3 Behavioural dependence of SC saliency coding (fixation-aligned). A subset of the total fixations were extracted based on saccade-goal (that is, the direction of the next-saccade relative to the RF; top illustration). (a,b) Average normalized firing rate of SCs neurons associated with high (dark red) versus low (light red) saliency (upper versus lower tertiles), when next-saccades were directed in a versus b opposite the RF. (d,e) Average normalized firing rate of SCi neurons associated with high (dark blue) versus low (light blue) saliency, when next saccades were directed in d versus e opposite the RF. Shading along the response curves indicates ±1 s.e.m. Tick marks above the abscissa indicate significant differences between the response curves within the pre-saccadic epoch (5 ms bins, Wilcoxon paired-samples test, Bonferroni-Holm correction). The dotted vertical line in a,b and d indicates the time the response curves first differed. (c,f) Average normalized firing rate during the pre-saccadic epoch (0–200 ms) illustrated by the gray shaded regions in a,b,d and e. Error bars in c and f indicate ±1 s.e.m. *P<0.05, paired t-test, 1-tailed; NS=not statistically significant.
Figure 3 Behavioural dependence of SC saliency coding (fixation-aligned). A subset of the total fixations were extracted based on saccade-goal (that is, the direction of the next-saccade relative to the RF; top illustration). (a,b) Average normalized firing rate of SCs neurons associated with high (dark red) versus low (light red) saliency (upper versus lower tertiles), when next-saccades were directed in a versus b opposite the RF. (d,e) Average normalized firing rate of SCi neurons associated with high (dark blue) versus low (light blue) saliency, when next saccades were directed in d versus e opposite the RF. Shading along the response curves indicates ±1 s.e.m. Tick marks above the abscissa indicate significant differences between the response curves within the pre-saccadic epoch (5 ms bins, Wilcoxon paired-samples test, Bonferroni-Holm correction). The dotted vertical line in a,b and d indicates the time the response curves first differed. (c,f) Average normalized firing rate during the pre-saccadic epoch (0–200 ms) illustrated by the gray shaded regions in a,b,d and e. Error bars in c and f indicate ±1 s.e.m. *P<0.05, paired t-test, 1-tailed; NS=not statistically significant.CC BY · PubMed Central
Figure 4 Behavioural dependence of SC saliency coding (saccade-aligned). (a,b) Average normalized activation of SCs neurons associated with high (dark red) versus low (light red) saliency (upper versus lower tertiles), when next-saccades were directed in a versus b opposite the RF. (d,e) Average normalized activation of SCi neurons associated with high (dark blue) versus low (light blue) saliency, when next saccades were directed in d versus e opposite the RF. Shading along the response curves indicates ±1 s.e.m. Tick marks above the abscissa indicate significant differences between the response curves (5 ms bins, Wilcoxon paired-samples test, Bonferroni-Holm correction). (c,f) Average normalized activation within a pre-saccadic epoch (−150 to −50 ms) illustrated by the grey shaded regions in a,b,d and e). Error bars in c and f indicate ±1 s.e.m. *P<0.05, paired t-test; NS=not statistically significant.
Figure 4 Behavioural dependence of SC saliency coding (saccade-aligned). (a,b) Average normalized activation of SCs neurons associated with high (dark red) versus low (light red) saliency (upper versus lower tertiles), when next-saccades were directed in a versus b opposite the RF. (d,e) Average normalized activation of SCi neurons associated with high (dark blue) versus low (light blue) saliency, when next saccades were directed in d versus e opposite the RF. Shading along the response curves indicates ±1 s.e.m. Tick marks above the abscissa indicate significant differences between the response curves (5 ms bins, Wilcoxon paired-samples test, Bonferroni-Holm correction). (c,f) Average normalized activation within a pre-saccadic epoch (−150 to −50 ms) illustrated by the grey shaded regions in a,b,d and e). Error bars in c and f indicate ±1 s.e.m. *P<0.05, paired t-test; NS=not statistically significant.CC BY · PubMed Central
Figure 5 Model saliency predicts saccade behaviour. (a) Number of saccades directed IN the RF as a function of model-predicted saliency. (b) Mean saccade reaction time (RT; that is, fixation duration) for saccades directed IN the RF as a function of model-predicted saliency. *P<0.05, Wilcoxon independent-samples test.
Figure 5 Model saliency predicts saccade behaviour. (a) Number of saccades directed IN the RF as a function of model-predicted saliency. (b) Mean saccade reaction time (RT; that is, fixation duration) for saccades directed IN the RF as a function of model-predicted saliency. *P<0.05, Wilcoxon independent-samples test.CC BY · PubMed Central
Figure 6 Feature dependence of SC saliency coding. Feature-specific activation profiles of SCs (red; n=34) and SCi (blue; n=26) neurons when the saccade-goal was in (a) versus opposite (b) the RF. Each column represents a given neuron, and the intensity of the colour represents the difference in activation evoked by high versus low model-predicted feature contrast (colour scale on the left). The table of numbers on the right indicates the mean across neurons for a given feature, with the bolded numbers/asterisk indicating the significant features. *P<0.05, Bonferroni-corrected t-tests, 1-tailed. Abbreviations: Lum: luminance; R–G: red–green colour opponency; B–Y: blue–yellow colour opponency.
Figure 6 Feature dependence of SC saliency coding. Feature-specific activation profiles of SCs (red; n=34) and SCi (blue; n=26) neurons when the saccade-goal was in (a) versus opposite (b) the RF. Each column represents a given neuron, and the intensity of the colour represents the difference in activation evoked by high versus low model-predicted feature contrast (colour scale on the left). The table of numbers on the right indicates the mean across neurons for a given feature, with the bolded numbers/asterisk indicating the significant features. *P<0.05, Bonferroni-corrected t-tests, 1-tailed. Abbreviations: Lum: luminance; R–G: red–green colour opponency; B–Y: blue–yellow colour opponency.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 56 references

Research works printed but not linked: 2

  1. White, B. J. & Munoz, D. P. Oxford Handbook of Eye Movements 195–213Oxford University Press (2011).
  2. Itti, L. The ilab neuromorphic vision c++ toolkit: free tools for the next generation of vision algorithms. Neuromorphic Eng. 1, 10 (2004).

Cited by

Items 101–191 of 191

Metadata

Authors
White, B. J., Berg, D. J., Kan, J. Y., Marino, R. A., Itti, L., & Munoz, D. P.
Affiliations
  • Queen's University (CA)
  • IBM Research - Almaden (US)
  • University of Southern California (US)
Corresponding author
White, Brian J.
Venue
Nature Communications
Published
2017-01-24
Volume
8
Issue
1
Pages
14263-14263
ISSN
2041-1723 (ISSN-L)
OpenAlex
W2580778720
Cited by
191
References
56
Citations by year
2026: 12 · 2025: 24 · 2024: 16 · 2023: 13 · 2022: 14 · 2021: 16 · 2020: 47 · 2019: 22 · 2018: 18 · 2017: 8
FWCI
10.78
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Science Foundation (1317433, 0827764, 1545089)
  • Canadian Institutes of Health Research (90910_1, 136972_1)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
Keywords
superior colliculus, visual saliency, saliency maps, visual attention, gaze control, primate vision, frontoparietal cortex
MeSH terms
Animals, Attention, Computer Simulation, Macaca mulatta, Male, Models, Neurological, Neural Pathways, Neurons, Photic Stimulation, Saccades, Software, Superior Colliculi, Thalamus, Visual Perception, Models, Animal

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.