DAQUAR Real-World Image Question Answering Dataset

DAQUAR, short for DAtaset for QUestion Answering on Real-world images, is a dataset for human question answering on images. The images in this dataset come from the NYU-Depth v2 dataset, all of which are RGBD images of indoor scenes, of which 795 are used for training and 654 are used for testing. There are two main types of question/answer pairs in DAQUAR: automatically generated and manually annotated.