arXiv subjects

Vasha DuTell

Publications and source records attributed to Vasha DuTell.

3 recordsLinked to original sources

arXiv2026

A Framework for Generating Semantically Ambiguous Images to Probe Human and Machine Perception

The classic duck-rabbit illusion reveals that when visual evidence is ambiguous, the human brain must decide what it sees. But where exactly do human observers draw the line between ''duck'' and ''rabbit'', and do machine classifiers draw it in the same place? We use semantically ambiguous images as interpretability probes to expose how vision models represent the boundaries between concepts. We present a psychophysically-informed framework that interpolates between concepts in the CLIP embedding space to generate continuous spectra of ambiguous images, allowing us to precisely measure where and how humans and machine classifiers place their semantic boundaries. Using this framework, we show that machine classifiers are more biased towards seeing ''rabbit'', whereas humans are more aligned with the CLIP embedding used for synthesis, and the guidance scale seems to affect human sensitivity more strongly than machine classifiers. Our framework demonstrates how controlled ambiguity can serve as a diagnostic tool to bridge the gap between human psychophysical analysis, image classification, and generative image models, offering insight into human-model alignment, robustness, model interpretability, and image synthesis methods.

cs.CV↗

arXiv2024

Seeing Faces in Things: A Model and Dataset for Pareidolia

The human visual system is well-tuned to detect faces of all shapes and sizes. While this brings obvious survival advantages, such as a better chance of spotting unknown predators in the bush, it also leads to spurious face detections. ``Face pareidolia'' describes the perception of face-like structure among otherwise random stimuli: seeing faces in coffee stains or clouds in the sky. In this paper, we study face pareidolia from a computer vision perspective. We present an image dataset of ``Faces in Things'', consisting of five thousand web images with human-annotated pareidolic faces. Using this dataset, we examine the extent to which a state-of-the-art human face detector exhibits pareidolia, and find a significant behavioral gap between humans and machines. We find that the evolutionary need for humans to detect animal faces, as well as human faces, may explain some of this gap. Finally, we propose a simple statistical model of pareidolia in images. Through studies on human subjects and our pareidolic face detectors we confirm a key prediction of our model regarding what image conditions are most likely to induce pareidolia. Dataset and Website: https://aka.ms/faces-in-things

cs.CV↗

arXiv2021

Efficient Dataflow Modeling of Peripheral Encoding in the Human Visual System

Computer graphics seeks to deliver compelling images, generated within a computing budget, targeted at a specific display device, and ultimately viewed by an individual user. The foveated nature of human vision offers an opportunity to efficiently allocate computation and compression to appropriate areas of the viewer's visual field, especially with the rise of high resolution and wide field-of-view display devices. However, while the ongoing study of foveal vision is advanced, much less is known about how humans process imagery in the periphery of their vision -- which comprises, at any given moment, the vast majority of the pixels in the image. We advance computational models for peripheral vision aimed toward their eventual use in computer graphics. In particular, we present a dataflow computational model of peripheral encoding that is more efficient than prior pooling - based methods and more compact than contrast sensitivity-based methods. Further, we account for the explicit encoding of "end stopped" features in the image, which was missing from previous methods. Finally, we evaluate our model in the context of perception of textures in the periphery. Our improved peripheral encoding may simplify development and testing of more sophisticated, complete models in more robust and realistic settings relevant to computer graphics.

cs.GR↗