Icelandic Vision Lab

@icevislab.bsky.social

1125 Followers 769 Following

Icelandic Vision Lab focuses on all things visual with a major emphasis on higher-level or "cognitive" aspects of vision: visionlab.is This is a joint account run by IVL PIs (Ásgeirsson, Hansmann-Roth, Kristjánsson, Ólafsdóttir, Sigurdardottir).

Shower gone wrong
Sheep
Cross-tabulation by percentiles: we split both VVIQ and VIVAS into five percentile bands based on each score’s position within the sample: 0-10%, 10-25%, 25-75% (middle half; IQR: interquartile range), 75-90%, and 90-100%; equal scores stay in the same band (ties kept together).
VIVAS color saturation distributions show that food imagery is particularly colorful, while imagery for novel objects is less vivid than for other categories.
VIVAS dimensions: Opacity, sharpness, color saturation
Photo of an eye by Beel coor on Unsplash: https://unsplash.com/photos/brown-and-black-eye-illustration-1AIHIjtuNCI
Black socks
Gabor patch earring
Shocked Pikachu Ishihara Plate
"They pay me in woims"
Panel A: Partial correlation plot between face discrimination ability and average discrimination ability for stubby animate-looking birds and hands; axes show residuals after partialling out average discrimination ability for spiky animate-looking birds and hands. Panel B: Partial correlation plot between face discrimination ability and average discrimination ability for spiky animate-looking birds and hands; axes show residuals after partialling out average discrimination ability for stubby animate-looking birds and hands (targets per second).
Multidimensional scaling was performed on the correlation matrix for performance (targets per second) on individual trials. This was done separately for (a) trials with novel objects and object-matched faces and (b) trials with familiar objects and object-matched faces. Each dot depicts a single trial, and the distance between two points is approximately equal to their dissimilarity. Trials are clustered based on our preregistered analysis pipeline: The faces cluster corresponds to trials contributing to the face discrimination dependent variable in Table 2, the stubby-animate objects cluster corresponds to trials contributing to factors added at Step 4 in Table 2, and the other objects cluster corresponds to trials contributing to factors added at Step 3 in Table 2. Dots are connected to their cluster’s barycenter via lines. Confidence ellipses for barycenters of the three clusters are shown, both for the 95% confidence level (light ellipses) and the 50% confidence level (dark ellipses).
O-factor in object perception
Stimuli belonged to one of four quadrants of object space. They were used in a visual foraging task where people searched for targets among distractors.
Object space coordinates of stimuli (11 exemplars of each type). Quadrants of object space are represented by different colors (green: spiky inanimate-looking; pink: stubby inanimate-looking; purple: spiky animate-looking; orange: stubby animate-looking; gray: all objects). Each dot represents object space coordinates of a particular exemplar object image. Exerc. eq. = exercise equipment.
The image demonstrates our overall approach to defining a visual object space. The depicted objects, network structure, activation patterns, and object space are purely illustrative. A convolutional neural network (VGG-16bn; Simonyan & Zisserman, 2014; see also Muttenthaler & Hebart, 2021) pretrained on ecologically relevant image categories (ecoset; Mehrer et al., 2017, 2021) was used to extract deep layer activations for a reference imageset (large and small animate and inanimate objects; Konkle & Caramazza, 2013). Principal components analysis (PCA) on these activations was used to define an object space that should capture the main two dimensions on which the reference objects vary in appearance. Network activation patterns of other objects, not used to construct object space, were then projected to this object space to extract their coordinates. A subset of these objects was used in the experiment.
Figure from Bao, She, McGill, & Tsao, 2020. Example images from the quadrants of object space.
Topographical organization of the high-level ventral visual stream.
Left: Big Five personality factors. Right: Big X (unknown) of object perception.
Example stimuli of 15 object types are shown. Quadrants of object space are represented by different colors (orange: stubby animate-looking; purple: spiky animate-looking; pink: stubby inanimate-looking; green: spiky inanimate-looking; gray: all objects). Exerc. eq. = exercise equipment.
Northern lights over Iceland
every day i get emails
The experimental design. On each trial, a sample image (face, house, or pseudoword) appeared in the middle of the screen, followed by match and foil images concurrently presented to the left and right of the center. The match was identical to the sample image; however, the foil was either featurally or configurally different. Match and foil pictures remained onscreen until the participant identified which image was identical to the sample image by pressing a button.
Examples of featural or configural changes of faces, houses, and pseudowords used in the visual recognition task.
Visual working memory task. Four randomly oriented bars were displayed for 1000 ms, followed by a 1000 ms mask. Subsequently, one of the items in the array was probed by a line serving as a location post-cue and the participants were asked to rotate it to indicate the remembered orientation. Arrows illustrate orientation adjustment and were not actually shown around the adjustment target.
Network plot of recognition accuracy in the different conditions of the visual object recognition task and visual working memory (VWM) precision for simple stimuli. The network plot shows the pattern in Pearson’s correlations between the six object memory conditions and VWM precision. The color of the connections indicates the strength and direction of the correlations, with strong positive correlations shown in red shades and weak positive correlations in blue shade; there were no negative correlations. Thicker arcs represent stronger correlations. The graph's structure, obtained through multidimensional clustering, simplifies the high-dimensional correlation data for easier interpretation. Configural house recognition shows the strongest positive correlation with VWM precision, indicated by the dark red arc.