Comparisons in performance between standard computational models and human observers with regard to scales and ISIs.

These results were averaged from four target categories except the sport category due to the special categorization of scale levels for this category. Two standard computational models SALIENCY and GIST that are believed to account for information conveyed during the first feedforward sweep could not reach human performance. Faster R-CNN built on backbone with FPN and ResNet-50, showed competitive recognition accuracy to human performance.