Object Recognition from Dynamic Cues: Examining the Inversion Effect

Humans can accurately categorize objects in dynamic scenes by relying on both static and dynamic cues. Inversion hinders recognition of static faces, static bodies, and dynamic body movements, but it remains unclear whether this effect generalizes to other categories, particularly when only dynamic cues are available. Motion cues from videos of four animate and four inanimate object categories were converted into object kinematograms and presented upright or inverted in an online experiment. For each video, participants were asked to: 1) determine if the item in the video was an animal or not, 2) determine its basic-level category, and 3) name it. We also provided an “I don’t know” option for all three questions. Overall, the likelihood of choosing the “I don’t know” response was higher for inverted than upright stimuli across all three questions. After removing the “I don’t know” responses, the animate category showed a significant inversion effect, while the inversion effect for the inanimate category was not significant. For the second question, we found the inversion effect to be variable across categories. While most animate categories (mammal, human, bird) showed significant inversion effects, for reptiles we failed to find an inversion effect. Surprisingly, some inanimate categories (vehicle, ball, pendulum) also showed a significant inversion effect. These differences might be related to variability in typical object orientation or motion pattern symmetry across categories. These results showcase the orientation-dependence of object categorization based on dynamic cues and inform future models of object recognition that apply better to dynamic environments.