Development and validation of an AI-generated real-world object stimuli set

Area of Science:

• Cognitive Psychology
• Computer Vision
• Neuroscience

Background:

• Manual curation of object stimuli for visual cognition research presents challenges due to feature variability.
• Existing datasets often lack control over stimulus parameters like size, color, and orientation.
• Artificial intelligence (AI) offers potential for generating highly realistic, customizable visual stimuli.

Purpose of the Study:

• To generate and validate a dataset of 200 AI-created images of everyday objects for research use.
• To provide stimuli suitable for embedding in virtual scenes and rendering in various colors.
• To assess the nameability, perceived realism, and familiarity of AI-generated stimuli across different age groups.

Main Methods:

• Utilized AI to generate 200 realistic images of everyday objects, oriented on a flat surface.
• Images were created in greyscale, allowing for flexible color rendering.
• Conducted a validation study with 90 adults (younger and older) assessing nameability, realism, and familiarity.

Main Results:

• The majority of AI-generated stimuli received high ratings for nameability, realism, and familiarity.
• No significant age-related differences were found in the perception of the stimuli.
• The dataset includes stimuli in seven colors and their validation scores, along with image statistics.

Conclusions:

• The AI-generated object stimuli are validated and suitable for diverse research applications in visual cognition.
• The open availability of this dataset facilitates future research in the field.
• AI advancements enable the creation of tailored visual stimuli, overcoming limitations of traditional datasets.