AI Models Gain Peripheral Vision But Still Lag Humans in Detecting Objects
-
Researchers created a dataset to simulate peripheral vision in AI models, which improved the models' ability to detect objects in the visual periphery.
-
However, the AI models still performed worse than humans at detecting objects, especially in the far periphery.
-
Unlike humans, factors like object size and visual clutter didn't impact the models' performance much.
-
The models may not be using context the same way humans do for visual detection tasks.
-
Understanding differences between human and AI peripheral vision could enable systems to better predict human behavior, like hazard detection for drivers.