SATBench

SATBench是由纽约大学创建的大规模数据集，专注于人类和动态神经网络在物体识别中的速度-准确性权衡（SAT）。该数据集包含148名观察者和4种神经网络在8个任务上的表现，主要用于研究在不同反应时间下的人类物体识别性能。数据集通过一系列实验收集，其中每个实验涉及不同的反应时间，以观察人类和神经网络的准确性如何随时间变化，从而为理解人类和机器视觉提供关键数据。

SATBench is a large-scale dataset developed by New York University, focusing on the speed-accuracy tradeoff (SAT) in object recognition between humans and dynamic neural networks. This dataset includes performance data from 148 human observers and 4 types of neural networks across 8 tasks, and is primarily used to study human object recognition performance under varying reaction times. The dataset is collected through a series of experiments, each involving distinct reaction times, to observe how the accuracy of humans and neural networks changes over time, thus providing critical data for understanding both human and machine vision.