Reverse-engineer the Distributional Structure of Infant Egocentric Views for Training Generalizable Image Classifiers
This addresses the problem of training generalizable image classifiers by leveraging infant visual data, offering a novel data augmentation approach with potential broad applications.
The paper analyzed infant egocentric views of attended objects, finding they have more diverse distributions than adult views, and demonstrated that computationally simulating this distribution improves generalization in image classifiers for both infant egocentric and third-person computer vision.
We analyze egocentric views of attended objects from infants. This paper shows 1) empirical evidence that children's egocentric views have more diverse distributions compared to adults' views, 2) we can computationally simulate the infants' distribution, and 3) the distribution is beneficial for training more generalized image classifiers not only for infant egocentric vision but for third-person computer vision.