DUT stands for Device Under Test. It is an image dataset used to evaluate salient object detection tasks. The dataset contains 5,168 high-definition images with a maximum side length of 400 pixels. These images have one or more salient objects and relatively complex backgrounds.

So far, this dataset is the only large-scale real-annotated dataset that includes eye fixations, bounding box, and pixel-wise ground truth. The images in it are difficult and more challenging, providing greater room for improvement in related research on salient object detection.