|Description (include details on usage, files and paper references)||The dynamic scenes dataset contains image sequences consisting of overall 1936 images. The images are taken from a camera inside a driving car and mainly show rural roads with high dynamics of driving vehicles at an image resolution of 752×480 pixels. Cars appear at all scales from as small as 15 pixels up to 200 pixels. The database consists of 176 sequences with 11 successive images each. It is split into equal size training and test sets of 968 images.
To evaluate pixel level labeling accuracy the last frame of each sequence is labeled pixel-wise, while the remainder only contains bounding box annotations for the frontal and rear view car object class. Overall, the dataset contains the
eight labels void, sky, road, lane marking, building, trees & bushes, grass and car.
A Dynamic Conditional Random Field Model for Joint Labeling of Object and Scene Classes
Christian Wojek, Bernt Schiele, ECCV 2008