@IceVisLab

Icelandic Vision Lab focuses on all things visual with a major emphasis on higher-level or “cognitive” aspects of vision. BlueSky: https://xitter.cf/t.co/O1mKxI4ydz

 Iceland
visionlab.is
Joined May 2018
  • Tweets3.7K
  • Following1.6K
  • Followers2.4K
Conversely, when appearance and semantics clash (objects on the “incorrect” side of object space: animate-looking inanimate objects, inanimate-looking animals), visual information is underutilized or misleading, and semantic relatedness hinders discrimination. 5/7
1
1
50
In a series of studies, we show that when objects' appearance matches their true superordinate category (objects on the correct side of object space), people effectively use diverse visual aspects for their discrimination, with minimal interference from semantic relatedness. 4/7
1
1
52
We reasoned that if such an object space exists in humans, it should be reflected in people’s object perception. We used CNNs trained for object classification to construct a visual object space, and projected images onto this space to pick objects with specific coordinates. 3/7
1
1
61
Thank you all for #VSS2024 @VSSMtg -- see you next year #VSS2025 #VSS – at St Pete Beach, FL
Club Vision at VSS

ALT Club Vision at VSS

1
21
1,532
Hi all, great to be back at @VSSMtg Here is stuff from Icelandic Vision Lab members from this afternoon and onward #VSS #VSS2024 – at St Pete Beach, FL
4
17
1,113
December has been a busy time at the Icelandic Vision Lab, and we now have two new doctors: Dr. Christian Houborg @CHouborg and Dr. Bahareh Jozranjbar @Bahareh32647110 It was great to have you with us, congratulations again, and see you later 😊 – at Iceland
Ph.D. defense photos

ALT Ph.D. defense photos

1
10
828
The Icelandic Vision Lab proudly presents our newest Ph.D. -- DOCTOR Sofia Krasovskaya! 🥳😎👏 – at Iceland
Dr. Sofia Krasovskaya

ALT Dr. Sofia Krasovskaya

1
26
1,760
Face discrimination tapped into mechanisms that supported other object discrimination. Specifically, people who struggled with telling apart faces also had difficulties with discriminating other objects from the face quadrant of object space, both novel and familiar. 7/11
Clustering of individual differences in visual foraging. 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 faces matched to objects, and b) trials with familiar objects and faces matched to objects. Each dot depicts a single trial and the distance between two points is approximately equal to their dissimilarity. Confidence ellipses for barycenters of three clusters are shown, both for the 95% confidence level (light ellipses) and the 50% confidence level (dark ellipses).

ALT Clustering of individual differences in visual foraging. 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 faces matched to objects, and b) trials with familiar objects and faces matched to objects. Each dot depicts a single trial and the distance between two points is approximately equal to their dissimilarity. Confidence ellipses for barycenters of three clusters are shown, both for the 95% confidence level (light ellipses) and the 50% confidence level (dark ellipses).

1
169
We administered a visual foraging task (N=511) where faces and other objects had to be discriminated. Stimuli were carefully picked out of tens of thousands of images to either match or not match with the coordinates of faces in object space. 6/11
Visual foraging stimuli and their object space coordinates. A. 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. B. 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.

ALT Visual foraging stimuli and their object space coordinates. A. 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. B. 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.

1
175
We utilized a deep convolutional neural network to construct a visual object space extracted from deep layer activations for a reference set of objects. The first two dimensions of object space capture attributes that distinguish one object from another. 3/11
The image demonstrates our overall approach to defining a visual object space. A convolutional neural network pretrained on ecologically relevant image categories was used to extract deep layer activations for a reference image set (large and small animate and inanimate objects). 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 were then projected to this object space to extract their coordinates.

ALT The image demonstrates our overall approach to defining a visual object space. A convolutional neural network pretrained on ecologically relevant image categories was used to extract deep layer activations for a reference image set (large and small animate and inanimate objects). 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 were then projected to this object space to extract their coordinates.

1
1
261
Preprint🚨 "Objects, faces, and spaces: Organizational principles of visual object perception as evidenced by individual differences in behavior" by @heidasigurdar and @IngaMaria86 This one only took *checks notes* two and a half years... A thread 🧵1/11 biorxiv.org/content/10.1101/…
15 types of objects, both novel and familiar, from the four quadrants of object space.

ALT 15 types of objects, both novel and familiar, from the four quadrants of object space.

2
21
1
48
8,495
Thank you @ecvp2023 #ecvp2023 for interacting with our work, the poster award (@PhilippeBlonde et al.), coming to our symposia and huge thanks to the presenters+ co-organizers, thanks for introducing us to your science, the fun conversations, dancing, and thanks for all the fish!
ECVP 2023

ALT ECVP 2023

4
26
2,180
6/ One key finding is the potential specificity of face recognition. It may rely on mechanisms partially distinct from those used in recognizing pseudowords and houses.
1
1
114
5/ The best-fitting statistical model included the time model, face model, and combination model. This was further supported by Bayesian model averaging. These predictors may play a role in producing the observed data patterns of what abilities in object recognition go together.
1
1
115
4/ Conceptual models were based on predicted patterns on how abilities in object recognition cluster together: category-specific model, process model, cost model, fine-tuning model, time model, face specialization model, word specialization model, and combination model.
1
2
123
2/ Our aim was to understand the principles of visual object processing and the relationship between abilities in recognition. Participants performed a delayed match-to-sample task with unfamiliar faces, houses, and words as stimuli, manipulated featurally and configurally.
1
1
246
Goodbye #VSS2023 As someone next to us at the airport said: "If DeSantis had his way, there would be no rainbow." We should think about the safety of our members. Equality for all. – at Tampa, FL
55
1,698
Load more