Refined Annotations with Soft Labels for Common Object Detection Benchmarks

These datasets contain refined annotations for four popular object detection benchmark datasets: COCO 2017 validation data, Pascal VOC 2012 detection data for both training and validation split, KITTI 2D object detection training data, and Cityscapes training and validation data. Each annotation consists of a bounding box (xmin, ymin, xmax, ymax) and a soft label annotation (\"postprocessed_softlabel\") as well as original annotator responses for object categories by annotator ID, including answer times.