Category-based attention facilitates the recognition of actions

Area of Science:

• Cognitive Psychology
• Computer Vision
• Neuroscience

Background:

• Action recognition is crucial for human interaction and AI.
• Efficiently prioritizing information sources is key to accurate action recognition.
• Category-based attention, a form of top-down processing, may play a significant role.

Purpose of the Study:

• To investigate how category-based attention influences the recognition of basic actions.
• To determine if attention facilitation effects differ between avatar and naturalistic images.
• To examine the impact of different cue types (word vs. image) on action recognition.

Main Methods:

• Four experiments were conducted using static avatar and naturalistic images.
• Participants identified basic actions (climb, jump, run, walk) after receiving valid or invalid category cues (word or avatar).
• Performance metrics included accuracy, reaction time (RT), and sensitivity (d-prime).

Main Results:

• Word cues facilitated action recognition for avatar stimuli (Experiments 1 & 2).
• Avatar cues aided recognition of naturalistic images, but primarily improved accuracy (Experiment 3).
• Word cues robustly enhanced both accuracy and RT for naturalistic images (Experiment 4).

Conclusions:

• Category-based attention enhances action recognition by pre-activating relevant representations.
• Top-down facilitation is preserved for naturalistic stimuli when using word cues.
• Attention mechanisms are vital for both human and artificial action recognition systems.