Journal of Vision最新文献

Visual motion perception and pursuit eye movements rely on integrating uncertain sensory input with prior knowledge. Previous work has extensively investigated motion perception biases from experience-related or innate priors. In parallel, since Eileen Kowler's pioneering work, anticipatory smooth eye movements have been studied as an indicator of cognitive expectations. However, whether perception and eye movements rely on the same priors and computational operations (e.g., Bayesian reliability-based integration) remains only partly understood. Additionally, the role of natural directional biases in two-dimensional space (e.g., cardinal preferences) and their interaction with immediate motion expectations have not been explored. To address these questions, we measured smooth pursuit and direction estimations in human volunteers tracking random dot kinematograms with a proportion of coherent dots (5%, 15%, or 40%) moving in one of 16 directions between -180° and +180° across three sessions: one with uniformly distributed directions and two including a specific directional bias. Under high uncertainty, inaccurate direction estimations systematically avoided the most frequent direction in biased sessions, contrary to the Bayesian attraction-to-prior predictions, and generally favored cardinal directions. In contrast, eye movements agreed with the attraction-to-prior effect: Anticipatory pursuit roughly aligned with expected directions, and early pursuit acceleration was enhanced when stimulus direction matched expectation. These findings highlight a dissociation between perception and pursuit in directional biases induced across time scales in two-dimensional space. This suggests that the two systems either rely on partly different internal models or use shared priors differently, pointing to a layered, task-dependent organization of motion inference in the brain.

Contrast modulation (CM) stimuli have been previously used to reveal nonlinear contributions of Y-like retinal ganglion cells (RGCs) such as parasol cells to cortical responses and perception. To test whether CMs are selectively processed within the magnocellular pathway, we assessed envelope motion discrimination and detection for achromatic (yellow-black) and chromatic (red-green) CMs in the presence of luminance masking noise to disrupt luminance-based mechanisms of motion processing. Compared to achromatic CMs, perception of chromatic CMs was more sensitive to luminance masking noise, suggesting that CM envelope motion perception relied predominantly on luminance signals. Specifically, envelope motion discrimination performance was better maintained for achromatic CMs than chromatic CMs, even at high masking noise levels. Notably, luminance masking noise greatly impaired envelope direction discrimination for chromatic CMs but had minimal impact on their detection, suggesting that chromatic aberrations may enhance envelope motion perception for chromatic CMs by introducing luminance signals. These findings reinforce previous neurophysiological and psychophysical evidence that CM stimuli selectively engage the nonlinear receptive field mechanisms of Y-like/parasol RGCs within the magnocellular retinogeniculate pathway, underscoring their potential to specifically target this pathway.

When two probes are flashed at different times within a moving frame, they can be perceived as dramatically separated from each other even though they are at the same location in the display. This effect suggests that we perceive object position relative to the surrounding frame even when it is moving (Özkan et al., 2021). Here, eight experiments reveal new properties of this frame effect. First, the influence of the frame on the perceived probe positions extends beyond its bounding contours by several degrees of visual angle, both in the direction of the frame's motion and orthogonal to it. It is also undiminished when the probes and the frame are in different depth planes. However, the influence of the frame's motion shows no extension in time-there is no effect on probes presented after the frame is removed and none retroactively before the frame appears either. The frame effect is also driven primarily by the displacement of the frame, not by its motion signals: the effect is stronger for moving bounded frames compared with moving unbounded random dot textures. When the bounded region has an internal texture that moves with or against the frame's motion or remains static, it is the displacement of the frame that produces the perceived position shifts of the probes, and the effect of the internal motion is mostly suppressed. The frame's influence is unaffected by whether the motion is self-initiated or not and does not diminish in strength across 2 hours of testing.

It remains an open question to what extent current deep neural networks (DNNs) are suitable computational models of the human visual system. While DNNs have proven to be capable of predicting neural activations in primate visual cortex with great success, psychophysical experiments have shown behavioral differences between DNNs and human observers. One of these behavioral differences is which individual images DNNs and human observers find easy or difficult to recognize, as quantified by error consistency (EC). Hypothetically, the reported differences in EC could arise late in visual processing, even though the representations extracted by DNNs and human observers may have been more similar in the initial forward sweep: At the presentation and response times investigated in earlier work, observer-internal idiosyncrasies (e.g., in feedback-mediated memory) might have influenced the final behavioral responses, lowering EC between DNNs and human observers. To test this hypothesis, we systematically vary presentation times of backward-masked stimuli from 8.3 to 267 ms and measure human performance on a speeded eightfold identification task with natural images. Contrary to the hypothesis that error consistency peaks early in time, we find that it never exceeds the value of 0.4 known from previous work with longer presentation times, suggesting that the differences between DNNs and humans cannot be explained by late high-level reasoning but point to systematic processing differences between DNNs and the early human visual system.

In order to fully process items of interest, we use information from outside the fovea to plan the target of the next saccadic eye movement. There is growing evidence that our initial preview of the identity and features of the saccade target, prior to bringing it to the fovea using the saccade, is used to make our subsequent post-saccadic processing more efficient. However, the mechanisms underlying trans-saccadic previews remain unknown. We investigated this in a gaze-contingent preview paradigm in which a face stimulus either remained the same ("valid preview") or changed ("invalid preview") during the saccadic eye movement. On some trials, a brief blank gap was added at the beginning of the new fixation, before the face was presented at the fovea. Although the expected preview benefit was found when the face stimulus was present after the saccade, the addition of the blank period eliminated the preview effect. Our results suggest that the preview effect relies on a sensorimotor prediction about both "what" will be present at the fovea after the saccade and "when" the new fixation will begin. These findings provided further evidence for an active, predictive mechanism for trans-saccadic perception.

Previous research has shown that observers can make reliable judgments about the relative mass of moving objects that collide in animated displays. One popular explanation of this is that observers' judgments are based on an internal model of Newtonian dynamics. An alternative explanation is that these judgments are based on measurable optical properties that are correlated with relative mass. To better understand this issue, the present investigation reanalyzed the data from three previous studies by Mitko and Fischer (2023), Sanborn et al. (2013), and Todd and Warren (1982), and it replicated an additional study by Hamrick et al. (2016). These new analyses demonstrate that observers' judgments of relative mass are most likely based on the post-collision optical velocities of objects without having to invoke an implausible mental representation of Newtonian dynamics as has been argued by several previous investigators.

The low prevalence effect, which posits that people are more likely to miss a present target when its prevalence rate is low, has important implications for real-world scenarios such as cancer screening and bomb detection. This effect has primarily been studied under full visibility; however, real-world scenarios often come with incomplete visibility. Occlusion and poor visibility introduce perceptual uncertainty, potentially altering how people decide whether a target is present. Here, we applied Bayesian decision theory to a visual search paradigm with partial occlusion, examining how target prevalence (prior) and the degree of occlusion (likelihood information) affect search decisions. Participants made target present/absent responses to target/distractor stimuli. In Experiment 1, all items were invisible, forcing participants to rely on trial feedback to learn the target's prevalence. Experiment 2 also provided trial feedback, but allowed either a small or large portion of the display to be visible. Target prevalence varied between blocks (high, 50%; low, 25%). Results showed that, when items were entirely hidden, participants learned to probability match the target's prevalence. However, when some items were visible, participants rarely responded present when the target was in the occluded region. Comparing the data with models (e.g., probability matching, Bayesian maximizing) revealed mixed strategies. This study introduces a novel method for investigating visual search under occlusion and suggests that, although people integrate prevalence and sensory input, their decisions are not fully Bayesian.

As research on human behavior, such as spatial navigation, increasingly adopts naturalistic settings, establishing best practices for such experiments becomes essential. Although virtual reality offers a bridge between laboratory control and real-world complexity, it does not fully capture the experiential richness of real-world environments. Here, we present a demonstration of a mobile eye-tracking study conducted in a large-scale, outdoor urban environment, featuring unconstrained, long-duration free exploration and outside-pointing tasks. Using the city of Limassol, Cyprus, as our testbed, we showcase the feasibility of collecting high-quality mobile eye-tracking, head orientation, and GPS data "in the wild," capturing a wide range of natural behavior with minimal experimental constraints. Based on this experience, we provide a set of best practices tailored to the logistical and methodological challenges posed by complex, real-world urban settings, challenges unlikely to arise in traditional indoor or highly controlled environments. Although these recommendations have general relevance, we exemplify them in the context of spatial navigation research. By establishing methodological standards for studies at this scale, we aimed to encourage and inform future research into naturalistic human behavior outside the laboratory.

Eileen Kowler was a member of the team of cognitive scientists that was assembled over a half a century ago by Robert M. Steinman. This team had grown and spanned several institutions in this country, as well as in Europe. The mechanisms governing the movements of the eyes were always at the center of Eileen's attention, but her research, as well as the research of our interdisciplinary group, is best understood by realizing its roots in Gestalt psychology. In the first part, this paper describes perspectives of two of the authors from the point of view of Eileen Kowler's collaborators. The second part provides a new look at the origins of Gestalt psychology, pointing out the scientific context of what has been referred to as the Gestalt revolution. The Gestalt revolution closely followed similar revolutions in mathematics and in physics, and it gave rise to new questions and challenges that have received recognition only recently.

Lightness constancy, the ability to create perceptual representations that are strongly correlated with surface reflectance despite variations in lighting and context, is a challenging computational problem. Indeed, it has proven difficult to develop image-computable models of how human vision achieves a substantial degree of lightness constancy in complex scenes. Recently, convolutional neural networks have been developed that are proficient at estimating reflectance, but little is known about how they achieve this, or whether they are good models of human vision. We examined this question by training a convolutional neural network to estimate reflectance and illumination in a computer-rendered virtual world, and evaluating both the convolutional neural network and human observers in a lightness matching task. In several conditions, we eliminated cues potentially supporting lightness constancy: local contrast, shading, shadows, and all contextual cues. We found that the network achieved a high degree of lightness constancy, outperforming human observers. However, we also found that eliminating cues affected the convolutional neural network and humans very differently. Humans were most affected when local contrast cues were made uninformative, whereas the convolutional neural network mostly relied on shading and shadows. In a follow-up experiment, we found that the convolutional neural network could learn to exploit noise artifacts typically associated with ray tracing and correlated with illuminance, with potential implications for the many studies relying on ray-traced images. We conclude that convolutional neural networks can learn an effective, global strategy of estimating lightness, which is closer to an optimal strategy for the ensemble of scenes we studied than the computation used by human vision.