Visual search shifts weight up the visual hierarchy from attentional guidance to verification
Abstract
When searching for target objects among distractors, observers exploit information contained in the attentional template, a memory representation of target features that supports both search guidance and verification for target selection or distractor rejection. The good-enough framework proposes that, because guidance and verification serve distinct computational goals, the precision of the template varies within a given feature dimension: guidance relies on a coarser, relational code that optimizes the reduction of the search space to candidate target locations, whereas verification relies on a more precise, optimal code for target-identity decisions. However, in searches involving real-world, multidimensional stimuli, it remains unclear whether the representation in the template also varies along the depth of the visual hierarchy, from low-level visual features to more abstract information related to object identity. Here, we used a deep convolutional neural network (DCNN) to estimate target–distractor similarity at different levels of the visual hierarchy. We then manipulated this similarity in a target-present versus target-absent two-alternative forcedchoice (2AFC) task while recording eye movements. Guidance, operationalized as which of the two stimuli was fixated first and modeled as a function of their relative similarity to the target, was predicted primarily by low-level similarity, with a smaller contribution of high-level similarity. Verification, operationalized as dwell time on each distractor and modeled as a function of its similarity to the target, was also predicted by both low and high-level similarity, but showed a clear shift toward high-level similarity relative to guidance. Critically, the relative weight of high-level similarity nearly doubled from guidance to verification. Together, our findings reveal a shift in representational depth up the visual hierarchy. Beyond template precision, representational depth therefore constitutes a further axis along which template information is selectively exploited according to the computational demands of different epochs of visual search.
Related
- Eye movements reflect attentional navigation of number space without visual experience
- Attentional modulation of persistent visual representations after stimulus offset is coordinated across primary visual and anterior cingulate cortices
- Mapping the Representational Basis of Object-Based Guidance: How Colour Availability Shapes Contributions Across the Hierarchy
- An adaptive fractional state links circuit mechanisms to cortical dynamics across the visual hierarchy
- Decomposing deep neural network-brain representational similarity reveals distinct sources across the human visual hierarchy
BibTeX
@misc{massironi2026visual,
title = {{Visual search shifts weight up the visual hierarchy from attentional guidance to verification}},
author = {Andrea Massironi and Emanuela Bricolo and Marco Alessandro Petilli},
howpublished = {PsyArXiv},
year = {2026},
month = oct,
doi = {10.31234/osf.io/9ye3r_v1},
url = {https://osf.io/preprints/psyarxiv/9ye3r_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Visual search shifts weight up the visual hierarchy from attentional guidance to verification
AU - Andrea Massironi
AU - Emanuela Bricolo
AU - Marco Alessandro Petilli
PY - 2026
DA - 2026/10/09
PB - PsyArXiv
DO - 10.31234/osf.io/9ye3r_v1
UR - https://osf.io/preprints/psyarxiv/9ye3r_v1/
AB - When searching for target objects among distractors, observers exploit information contained in the attentional template, a memory representation of target features that supports both search guidance and verification for target selection or distractor rejection. The good-enough framework proposes that, because guidance and verification serve distinct computational goals, the precision of the template varies within a given feature dimension: guidance relies on a coarser, relational code that optimizes the reduction of the search space to candidate target locations, whereas verification relies on a more precise, optimal code for target-identity decisions. However, in searches involving real-world, multidimensional stimuli, it remains unclear whether the representation in the template also varies along the depth of the visual hierarchy, from low-level visual features to more abstract information related to object identity. Here, we used a deep convolutional neural network (DCNN) to estimate target–distractor similarity at different levels of the visual hierarchy. We then manipulated this similarity in a target-present versus target-absent two-alternative forcedchoice (2AFC) task while recording eye movements. Guidance, operationalized as which of the two stimuli was fixated first and modeled as a function of their relative similarity to the target, was predicted primarily by low-level similarity, with a smaller contribution of high-level similarity. Verification, operationalized as dwell time on each distractor and modeled as a function of its similarity to the target, was also predicted by both low and high-level similarity, but showed a clear shift toward high-level similarity relative to guidance. Critically, the relative weight of high-level similarity nearly doubled from guidance to verification. Together, our findings reveal a shift in representational depth up the visual hierarchy. Beyond template precision, representational depth therefore constitutes a further axis along which template information is selectively exploited according to the computational demands of different epochs of visual search.
ER -