PASCAL Context Semantic Segmentation Dataset

This dataset is a set of additional annotations for the PASCAL VOC 2010 dataset, which goes beyond the original PASCAL semantic segmentation task and provides annotations for the entire scene.

This dataset contains pixel-level annotations of all training images, with more than 400 categories (including the original 20 categories and background from PASCAL VOC segmentation), divided into three categories: objects, stuff, and hybrids. It has fewer object classes, so a subset of 59 common classes is usually selected for use. The statistics are the same as the original PASCAL dataset, with 10,103 images in the training and validation sets and 9,637 images in the test set.