Ultra-rapid object categorization in real-world scenes with top-down manipulations

Area of Science:

• Cognitive Psychology
• Computational Vision
• Neuroscience

Background:

• Visual object recognition is rapid and effortless, typically involving bottom-up and top-down cognitive processing.
• The role of top-down processing in ultra-rapid categorization (brief stimuli, limited response time) remains unclear.
• Feedforward information is often assumed sufficient for initial scene discrimination in ultra-rapid scenarios.

Purpose of the Study:

• To investigate the interaction of top-down manipulations (category level, type, real-world size) in ultra-rapid object categorization.
• To determine if top-down processing contributes to object recognition under time constraints.
• To compare human performance with computational models in ultra-rapid categorization tasks.

Main Methods:

• Developed a dataset of real-world scene images with measured target object display sizes.
• Measured human ultra-rapid object categorization performance using these images.
• Utilized standard feedforward computational models and a state-of-the-art object detection model for comparison.

Main Results:

• Ultra-rapid categorization was influenced by animacy (animal, vehicle, food), level of abstraction (people, sport), and real-world size.
• Evidence supports the involvement of top-down processing in rapidly categorizing specific objects, like 'sport' at a fine-grained level.
• Human and model comparisons revealed insights into potential collaboration and integration of cognitive processes.

Conclusions:

• Top-down processing plays a role even in ultra-rapid visual object categorization.
• Category level, type, and object size are significant factors influencing rapid recognition.
• Findings offer valuable perspectives for both experimental and computational vision research.