Dirk Bernhardt-Walther
LightNews LightNews
banner
dirkbwalther.bsky.social
Dirk Bernhardt-Walther
@dirkbwalther.bsky.social
540 followers 460 following 52 posts
Professor of Psychology University of Toronto, scene perception by humans and machines, visual aesthetics.
bwlab.org
Posts Replies Media Videos
dirkbwalther.bsky.social
Who's afraid of AI?

toronto2025.ai
October 16, 2025 at 7:07 PM
dirkbwalther.bsky.social
How do our senses contribute to food enjoyment?

We answer this question in a new review paper. The paper started as a final project in an undergraduate seminar on neuroaesthetics in 2023. I am so proud of the students and their hard work!

cjur.ca/wp-content/u...

@uoftpsychology.bsky.social
October 14, 2025 at 1:43 AM
dirkbwalther.bsky.social
What features guide visual search for complex scenes? In her new APP paper, @songaeun.bsky.social uses scenes generated with a GAN to isolate features that guide search: layout, material properties, and lighting.
rdcu.be/elPgO

@drmack.bsky.social
@macklab.bsky.social
@uoftpsychology.bsky.social
Target–distractor (T-D) Configurations and example search displays. A Target–distractor configurations. The black cell indicates targets, while the grey cells indicate the distractors in each condition. The areas in light grey indicate possible jitter zones for individual distractors. The exact position of the distractors can vary in different directions while maintaining the Euclidean distances from the target and configuration rule. B Example displays of search arrays. All three displays are the examples of target-present trials.
May 13, 2025 at 10:07 AM
dirkbwalther.bsky.social
Creativity as aesthetics in reverse? How does the Mirror Model of Art hold up when seen through the lens of neuroscience? It doesn't.

New paper with Oshin Vartanian, Delaram Farzanfar and Pablo Tinio in Neuropsychologia.
doi.org/10.1016/j.ne...

@uoftpsychology.bsky.social
@uoft.bsky.social
The Mirror Model of Art sees creativity and aesthetics as mirrors of one another.
April 11, 2025 at 1:05 AM
dirkbwalther.bsky.social
I invite you to attend the Toronto Aesthetics Sciences Conference (TASC) at @uc-uoft.bsky.social on April 8th, 2025.
Registration is free but required.
sites.google.com/view/tasc2025/

Please join us for a day of nerding out over empirical aesthetics!

@uoftpsychology.bsky.social @uoft.bsky.social
February 3, 2025 at 1:56 PM
dirkbwalther.bsky.social
Absolutely! We see this effect as a physiological explanation for the descending branch of the Wundt-Berlyne inverted U-shaped curve. Sufficient interest and arousal are still required! We literally talk about your ⬜ in the paper:
This proposal aligns with Berlyne's two-factor model (20), which posits that aesthetic appreciation is shaped by both arousal and cognitive effort, with preference increasing alongside arousal but decreasing with increased cognitive demand. This model explains why overly simple stimuli, such as a blank white room, fail to induce aesthetic pleasure despite requiring minimal cognitive effort—they lack sufficient arousal or interest.
January 29, 2025 at 10:02 PM
dirkbwalther.bsky.social
Energy efficiency drives evolution, and humans may have evolved pleasure-based signals to optimize actions. Does this extend to aesthetic pleasure?

Yes!

We find strong evidence in silico and human observers!
osf.io/preprints/ps...

With Yikai Tang and Wil Cunningham.
@uoftpsychology.bsky.social
Conceptual Roadmap of the present study. To examine the relationship between the metabolic costs of visual processing and aesthetic pleasure, we used both computational and physiological measures to quantify metabolic costs during visual processing: 1. Model-derived estimates of metabolic costs based on the activation of a deep neural network; 2. Metabolic activity of human brains, specifically in the visual processing areas. We found that both measures were inversely related to aesthetic pleasure.
January 29, 2025 at 7:43 PM
dirkbwalther.bsky.social
People like round contours better than angular contours.

Where and how is curvature represented in the brain?

With indoor architecture we found a neural dissociation between subjective curvature and computational curvature.

www.nature.com/articles/s41...
November 4, 2024 at 5:48 PM
dirkbwalther.bsky.social
How does neurodiversity affect aesthetic processing?

I explored this questions with students from a seminar class on Neuroaesthics.

Our insights just got published in Neuroscience and Behavioral Reviews.

authors.elsevier.com/a/1jlNnY3M3i...

I'm proud of the great work by the students!
September 11, 2024 at 3:16 PM
dirkbwalther.bsky.social
What is the role of curvature in aesthetic appeal? A star-studded symposium at #IAEA2024 explores this question in detail Find us at @IAEAPalma2024 on Thursday, May 9th, at 9 am, bright and early!
IAEA 2024 Symposium: The role of curvature in aesthetic appeal Thursday, May 9th, 2024, 9:00 – 10:30 am Lluís Domènech Auditòrium (lower ground floor) Moderators: Dirk B. Walther and Marco Bertamini Program: 9:00 – 9:05 Welcome 9:05 – 9:20 Nicole Ruta: Explaining the curvature effect 9:20 – 9:35 Letizia Palumbo: Preference for curvature: the role of individual differences and object attributes 9:35 – 9:50 Ronald Hübner: Individual differences in the preference for certain types of spirals 9:50 – 10:05 Marco Bertamini: The (lack of) effect of movement on visual aesthetic judgments 10:05 – 10:20 Dirk B. Walther: Neural Dissociation Between Computational and Perceived Measures of Curvature 10:20 – 10:30 General Discussion and Conclusion
May 4, 2024 at 11:56 PM
dirkbwalther.bsky.social
We are looking for candidates for a unique postdoc opportunity to work on the role of perceptual grouping in human and computer vision with Sven Dickinson, Kaleem Siddiqi, Zygmunt Pizlo and me.
www.bwlab.org/postdoc-posi...

Please repost.
Logo of the University of Toronto Logo of McGill University Logo of the University of Irvine
April 18, 2024 at 3:09 AM
dirkbwalther.bsky.social
We are looking for a new Lead Technologist for the Toronto Neuroimaging Facility (ToNI). Join us for the highly interesting, dynamic research and stay for the great benefits offered by the University of Toronto. To apply:
jobs.utoronto.ca/job/Toronto-...
Please repost.
Logo saying "Toronto Neuroimaging"
March 11, 2024 at 5:44 PM
dirkbwalther.bsky.social
pRF size decreases from anterior to posterior. Consistent with this gradient, preference for scenes versus faces also decreases from anterior to posterior.
doi.org/10.1093/cerc...
Average size of receptive field depicted within anatomical bounds of the ROIs. Average contrast coefficient for scenes versus faces depicted within the anatomical bounds of the ROIs.
January 18, 2024 at 6:00 PM
dirkbwalther.bsky.social
Nice feature on the UofT Arts and Science website about Dela Farzanfar's and my recent Psych Science article on manipulating aesthetic liking of images:
www.artsci.utoronto.ca/news/researc...
Link to the paper:
doi.org/10.3389/fcom...
January 16, 2024 at 4:19 PM
dirkbwalther.bsky.social
4/4

When we augment the input into the CNN with shape-based symmetry scores, categorization improves significantly. Shape features help with categorization but are not computed by the CNN.

Is this discrepancy due to CNN architecture or the training regime? We don’t know yet.
Top 1 and Top 5 performances in a 3-channel configuration on Places365. In each case, all layers of the network were trained from scratch. The top row shows the results of the traditional R,G,B input configuration, while the others show combinations of scene contours and scene contours weighted by our importance measures.
January 9, 2024 at 3:56 PM
dirkbwalther.bsky.social
3/4

Human subjects categorize scenes more accurately with the most than the least symmetric half-drawings. So does VGG16.
A comparison of human scene categorization performance with CNN performance. As with the human observer data, CNNs perform better on the high 50% split according to each importance measure, than the low 50% split.
January 9, 2024 at 3:52 PM
dirkbwalther.bsky.social
2/4

We developed algorithms to measure the amount of local parallelism, mirror symmetry, taper symmetry, and contour separation in drawings of scenes from the medial axis transform. We then separate the contours into the most and the least symmetric halves.

The code is available: mlvtoolbox.org
Examples of original photographs and the corresponding mirror symmetry, separation, parallelism, and taper scores on scene contours, using a jet colormap to show increasing values.
January 9, 2024 at 3:51 PM
dirkbwalther.bsky.social
Shape-Based Measures Improve Scene Categorization by DNNs.

Morteza Rezanejad's magnus opus finally out in IEEE PAMI. doi.org/10.1109/TPAM...

A super productive and fun collaboration with John Wilder, Allan D. Jepson, Sven Dickinson, and Kaleem Siddiqi.

Read on for a quick summary.
An illustration of our approach using an example from a database of line drawings of natural scenes. In this pipeline, we first extract line drawings of the input image, then its Medial Axis Transform (MAT) is computed and then the skeletal points on the MAT are scored based upon one of our importance score measures. Finally, they are projected back onto the contours. The figure on the top left shows an example scene used in our pipeline, and the figure below presents the jet colormap visualization of its contours using our medial axis-based contour importance measures.
January 9, 2024 at 3:50 PM
dirkbwalther.bsky.social
Good question. Yes, I think that's conceivable. But there has to be some topographic structure in all sections of the hippocampus for pRFs to even be fitted. We also see this organization in the systematic progression of eccentricity (Supp. Fig 1 - attached).
Average Eccentricity across segments of the hippocampus and surrounding medial temporal cortex regions (anterolateral and postero-medial entorhinal cortex, perirhinal cortex, and parahippocampal cortex).
November 25, 2023 at 3:06 PM
dirkbwalther.bsky.social
The hippocampus is not usually thought of as a visual brain region. But it shows a systematic progression of pRF size from anterior to posterior, which matches a similar gradient in scene versus face preference. New paper by Charlotte Leferink now out in Cerebral Cortex.
doi.org/10.1093/cerc...
Scenes versus faces gradient within the hippocampus, the parahippocampal cortex, the perirhinal cortex, and the entorhinal cortex overlaid on a coronal and two sagittal brain slices. Population receptive field size of voxels within the hippocampus, the parahippocampal cortex, the perirhinal cortex, and the entorhinal cortex overlaid on a coronal and two sagittal brain slices.
November 24, 2023 at 3:11 PM
dirkbwalther.bsky.social
A computational model of image complexity predicts complexity ratings, but the neural signatures of the two are completely different. New paper by Elizabeth Zhou, John Wilder, Claudia Damiano and me out in PACA today!
psycnet.apa.org/record/2024-...
PDF on our website:
www.bwlab.org/publications
Inflated brain images with annotated regions of interest showing neural activity related to complexity ratings, a computational model of image complexity, and their overlap. Model predictions show up in early visual cortex, ratings in higher level brain areas, with little overlap between the two.
November 21, 2023 at 3:19 PM
dirkbwalther.bsky.social
New paper from our lab by Seohee Han, Morteza Rezanejad and me on the memorability of line drawings out in Memory and Cognition.
Spoiler alert, T junctions are really important!
rdcu.be/dp2x5
Graph showing the feature importance for the memorability of photographs and line drawings of scenes. T junctions take the top spot.
November 1, 2023 at 7:37 PM
dirkbwalther.bsky.social
Trying this site for the first time to announce that our work with Claudia Damiano and Pinaki Gayen on visually communicating emotions in art is featured by
the U of T student newspaper The Varsity.
Here is the story:
thevarsity.ca/2023/10/15/t...
October 18, 2023 at 2:28 AM