Related Experiment Video
Updated: Feb 4, 2026

Improving 2D and 3D Skin In Vitro Models Using Macromolecular Crowding
Published on: August 22, 2016
A 2D Gabor-wavelet baseline model out-performs a 3D surface model in scene-responsive cortex
Anna Shafer-Skelton1,2,3, Timothy F Brady1, John T Serences1,4
1Department of Psychology, University of California, San Diego, La Jolla, California, United States of America.
A new study found that low-level visual features, like spatial frequency and orientation, better explain brain activity in scene-selective areas than 3D surface models. This challenges previous assumptions about how the brain processes complex spatial information.
Area of Science:
- Vision science
- Computational neuroscience
- Cognitive psychology
Background:
- Understanding 3D representations of spatial information in natural scenes is a challenge.
- Disentangling high-level 3D information from low-level features like spatial frequency and orientation is difficult.
- Model-comparison frameworks are used to analyze visual processing in scene-selective brain regions.
Purpose of the Study:
- To test if previous findings on 3D-surface features in scene-selective areas generalize to a new stimulus set.
- To investigate whether low-level visual features or 3D-surface features better explain neural responses in OPA, PPA, and MPA/RSC.
- To re-evaluate the role of "scene-selective" areas in processing 3D spatial information.
Main Methods:
- Used a model-comparison framework with novel stimuli designed to dissociate Gabor-wavelet features from 3D scene-surface features.
- Compared the explanatory power of a Gabor-wavelet baseline model against a 3D-surface model for voxel responses in specific visual cortex areas.
- Analyzed neural data from human participants viewing naturalistic scene stimuli.
Main Results:
- A Gabor-wavelet model (low-level, 2D) provided a better fit to voxel responses in OPA, PPA, and MPA/RSC than a 3D-surface model.
- Contrary to prior research, low-level visual features explained more variance in these scene-selective areas.
- The findings suggest that spatial frequency and orientation information may be more critical than previously thought for representing 3D scene properties.
Conclusions:
- The information processed in scene-selective regions might be predominantly in the form of low-level spatial frequency and orientation.
- Differences in baseline models across studies may explain conflicting results regarding 3D-surface representation.
- Further research is needed to disentangle low-level and high-level visual information processing and understand how real-world visual properties are cued.
Related Concept Videos
Response Surface Methodology
The process of RSM involves several key steps:
Molecular Models
The Bohr Model
The Fluid Mosaic Model
Responses to Drought and Flooding
Stereotype Content Model

