Are Deep Neural Networks Adequate Behavioral Models of Human Visual Perception?

Deep neural networks (DNNs), renowned for their breakthroughs in computer vision tasks like object classification and segmentation, are scrutinized as potential models for human visual perception. This article reviews evidence on DNNs' suitability as behavioral models for human core object recognition. After analyzing numerous studies on human and DNN object recognition performance, the article concludes that while DNNs are valuable scientific tools, they must be fully adequate computational models for human core object recognition. Several myths about DNNs in vision science are also debunked.