Global and high-level effects in crowding cannot be predicted by either high-dimensional pooling or target cueing

Abstract

In visual crowding, the perception of a target deteriorates in the presence of nearby flankers. Traditionally, target-flanker interactions have been considered as local, mostly deleterious, low-level, and feature specific, occurring when information is pooled along the visual processing hierarchy. Recently, a vast literature of high-level effects in crowding (grouping effects and face-holistic crowding in particular) led to a different understanding of crowding, as a global, complex, and multilevel phenomenon that cannot be captured or explained by simple pooling models. It was recently argued that these high-level effects may still be captured by more sophisticated pooling models, such as the Texture Tiling model (TTM). Unlike simple pooling models, the high-dimensional pooling stage of the TTM preserves rich information about a crowded stimulus and, in principle, this information may be sufficient to drive high-level and global aspects of crowding. In addition, it was proposed that grouping effects in crowding may be explained by post-perceptual target cueing. Here, we extensively tested the predictions of the TTM on the results of six different studies that highlighted high-level effects in crowding. Our results show that the TTM cannot explain any of these high-level effects, and that the behavior of the model is equivalent to a simple pooling model. In addition, we show that grouping effects in crowding cannot be predicted by post-perceptual factors, such as target cueing. Taken together, these results reinforce once more the idea that complex target-flanker interactions determine crowding and that crowding occurs at multiple levels of the visual hierarchy.

Figure 1.

Stimuli used to validate the TTM. In the original experiments, observers were asked to discriminate the offset of a vernier target presented in the right hemifield and in the periphery (here shown in the center of each image), while looking at a fixation dot. Different flanker configurations were presented across the studies: “Short/Same/Long lines” and “Boxes” in Manassi et al. (2012); “Completion” and “Butterflies” in Manassi et al. (2015); “Shapes” in Manassi et al. (2013); “Patterns” in Manassi et al. (2016). In the original experiments as well as in the TTM validations, the target eccentricity was 3.88 degrees in the “Lines,” “Boxes,” “Completion,” and “Butterflies” experiments, and 9 degrees in the “Shapes” and “Patterns” experiments. Note that, in all original experiments except “Patterns”, two vertical lines (pointers) were added above and below the vernier target to reduce target location uncertainty.

Figure 2.

Lines. Left. Data from Manassi et al. (2012). Offset discrimination thresholds were determined for vernier targets presented in the periphery at 4 degrees of eccentricity. Center. TTM validation in which observers discriminate between left and right offset verniers in mongrel images. Right. TTM validation with a template matching algorithm using the same mongrels as in the human experiment. Green dashed lines indicate vernier alone performance. Red lines indicate chance level (50% accuracy). Note that the y-axis labels are different.

Figure 3.

Completion. Left. Data from Manassi et al. (2015). Offset discrimination thresholds were determined for vernier targets presented in the periphery at 4 degrees of eccentricity. Center. TTM validation in which observers discriminate between left and right offset verniers in mongrel images. Right. TTM validation with a template matching algorithm using the same mongrels as in the human experiment. Note that the algorithm made 0% errors for in the comp2 condition (the data is not missing). Green dashed lines indicate vernier alone performance. Red lines indicate chance level (50% accuracy). Note that the y-axis labels are different.

Figure 4.

Boxes and crosses. Left. Data from Manassi et al. (2012). Offset discrimination thresholds were determined for vernier targets presented in the periphery at 4 degrees of eccentricity. Center. TTM validation in which observers discriminate between left and right offset verniers in mongrel images. Right. TTM validation with a template matching algorithm using the same mongrels as in the human experiment. Green dashed lines indicate vernier alone performance. Red lines indicate chance level (50% accuracy). Note that the y-axis labels are different.

Figure 5.

Shapes. Left. Data from Manassi et al. (2013). Offset discrimination thresholds were determined for vernier targets presented in the periphery at 9 degrees of eccentricity. Center. TTM validation in which observers discriminate between left and right offset verniers in mongrel images. Right. TTM validation with a template matching algorithm using the same mongrels as in the human experiment. Green dashed lines indicate vernier alone performance. Red lines indicate chance level (50% accuracy). Note that the y-axis labels are different.

Figure 6.

Patterns. Left. Data from Manassi et al. (2016). Offset discrimination thresholds were determined for vernier targets presented in the periphery at 9 degrees of eccentricity. Center. TTM validation in which observers discriminate between left and right offset verniers in mongrel images. Right. TTM validation with a template matching algorithm using the same mongrels as in the human experiment. Green dashed lines indicate vernier alone performance. Red lines indicate chance level (50% accuracy). Note that the y-axis labels are different.

Figure 7.

(A) TTM performance in the mongrel vernier offset discrimination task showed no correlation (r = −0.044, p = 0.799, BF₀₁ = 4.672; Ly, Verhagen, & Wagenmakers, 2016; Rouder, Speckman, Sun, Morey, & Iverson, 2009) with the original data from (Manassi et al., 2012; Manassi et al., 2013; Manassi et al., 2015; Manassi et al., 2016). (B) TTM performance as a function of the sum of the flanker pixels in the corresponding conditions. Each dot indicates a flanking condition in Figure 1. The red line indicates chance level performance. For illustrative reasons, we plotted all tested conditions in a unique graph. Separate plots for all experiments are shown in the supplementary information (Supplementary Information Figure SF). Fitting the data with a psychometric function (see Equation 3 in Supplementary Information SL), we found a strong correlation between the TTM and the fitted performance (r(34) = 0.796, p < 0.001, BF₁₀ > 10⁶).

Figure 8.

Right column, for both panels. Conditions in which the target location is weakly cued by the flanker configuration. Left Column, for both panels. Conditions in which the target location is strongly cued by the flanker configuration. If cueing had a strong impact on target discrimination performance, crowding would decrease from left to right in all comparisons. However, crowding strength either increases (left panel) or stays constant (right panel), while target cueing always increases. All conditions are taken from (Manassi et al., 2012; Manassi et al., 2013; Manassi et al., 2015; Manassi et al., 2016).

Figure 9.

Single face discrimination task. Observers were asked to discriminate which of the two images was a face (left or right, 2AFC), by pressing the left or right arrow, while fixating the central cross. Across the experiment, the face could be either upright or inverted. In these examples, an upright face is presented on the left side (left panel), and an inverted face is presented on the right side (right panel). Mooney faces reprinted from Schwiedrzik et al. (2018). Distributed under a CC-BY license.

Figure 10.

Examples of stimuli used in the face crowding task. There were three main conditions (upright target alone, target with upright flankers or target inverted flankers) presented at four different eccentricities. Mooney faces reprinted from Schwiedrzik et al. (2018). Distributed under a CC-BY license.

Figure 11.

TTM and single Mooney face discrimination. (A) Face discrimination task. Observers were asked to discriminate an upright/inverted face from a scrambled face at all tested eccentricities. Accuracy remained on a constant high level for all eccentricities. Crucially, accuracy was higher for upright than for inverted faces. (B) Mongrel face discrimination task. Accuracy decreased with increasing eccentricity, contrary to the behavioral results. Using a linear mixed effect model, no significant difference between the upright and inverted face conditions was observed (i.e., no significant effect of face orientation on model performance). Shaded regions indicate the standard error of the mean.

Figure 12.

TTM and crowding of Mooney faces. (A) Face crowding task, data from Farzin et al. (2009). Target discrimination performance decreased when eccentricity increased. When the target face was flanked by inverted faces, crowding increased with increasing eccentricity (orange). When the target was flanked by upright faces, crowding increased even more with eccentricity (blue). Shaded regions indicate the standard error of the mean. Stars indicate a significant difference in crowding strength between the upright and inverted flanker face conditions (paired Student t-test, 2-tails). (B) Mongrel face crowding task. Accuracy decreased with eccentricity. When analyzing the results using a linear mixed effect model, no effect of flanker face orientation was exposed. Shaded regions indicate the standard error of the mean.

Figure 13.

TTM mongrel examples used in the single face and gender face discrimination tasks. The stimuli (TTM input) are highlighted in red. To give a representative sample of the TTM outputs for each example, we show mongrels for different eccentricities. Note that we cropped the mongrels for ease of comparison. All mongrels can be found at https://github.com/albornet/TTM_Verniers_Faces_Mongrels. Mooney faces reprinted from Schwiedrzik et al. (2018). Distributed under a CC-BY license.