VADD is a large novel saliency dataset for autonomous driving containing 10342 training and 1601 validation images. The ground truth masks of the VADD dataset are obtained by taking advantage of semantic label information from publicly available driving scene datasets. These masks are superimposed on real images for saliency-heatmaps construction, highlighting on-road objects as salient, i.e., pedestrians, cyclists, cars, motorbikes, trucks, trams, traffic light, and traffic signs. Figure 1 illustrates the examples.
Example: Figure 1
Our CIAD Laboratory: CIAD
Our EPAN Research Group at UTBM: EPAN-UTBM