The PASCAL-Scirbble dataset, also known as ScribbleSup, is an extension of the PASCAL dataset with scribble annotations for semantic segmentation. The annotations follow two different protocols.

In the first protocol, the PASCAL VOC 2012 set (https://orion.hyper.ai/datasets/7660) with annotations, including 20 object categories (airplanes, bicycles, etc.) and one background category. There are 12,031 annotated images, including 10,582 images in the training set and 1,449 images in the validation set.

In the second protocol, the 59 object or content categories and one background category from the PASCAL-CONTEXT dataset were used. In addition to the 20 object categories from the first protocol, there were 39 additional categories (snow, trees, etc.). This protocol was followed to annotate the PASCAL-CONTEXT dataset. A total of 4,998 images from the training set were annotated.

The Chinese University of Hong Kong published this dataset at the 2016 IEEE CVPR