Scene context is predictive of unconstrained object similarity judgments

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Psychology

Background:

• Traditional models of object similarity focus on visual form or linguistic associations.
• Intuitive object grouping may rely on contextual reasoning, linking objects to similar scenes or events.

Purpose of the Study:

• To investigate the role of contextual reasoning in human object similarity judgments.
• To develop and evaluate a computational model based on scene context associations.

Main Methods:

• Large-scale analysis of natural scene statistics and human behavioral data.
• Utilized a convolutional neural network (CNN) to learn contextual prototypes from object-scene associations.
• Conducted behavioral experiments to compare model predictions with human similarity judgments.

Main Results:

• A computational model of object-scene context associations strongly predicted human object grouping.
• Contextual prototypes rivaled the predictive power of models based on visual CNN features or word embeddings.
• Natural scene statistics effectively capture commonsense notions of object similarity.

Conclusions:

• Contextual reasoning is a powerful predictor of human object similarity judgments.
• Computational models of scene context offer a novel approach to understanding object categorization.
• The findings highlight the importance of environmental statistics in shaping cognitive representations of objects.