FloodNet is dataset for post-flood scene understanding, i.e flood detection and distinguishing different water bodies and flood. It contains about 11, 000 question-image pairs for VQA and 3,200 images. It is introduced for image classification, semantic segmentation, and visual question answering.