Visual scene understanding is the core task in making any crucial decision in\nany computer vision system. Although popular computer vision datasets like\nCityscapes, MS-COCO, PASCAL provide good benchmarks for several tasks (e.g.\nimage classification, segmentation, object detection), these datasets are\nhardly suitable for post disaster damage assessments. On the other hand,\nexisting natural disaster datasets include mainly satellite imagery which have\nlow spatial resolution and a high revisit period. Therefore, they do not have a\nscope to provide quick and efficient damage assessment tasks. Unmanned Aerial\nVehicle(UAV) can effortlessly access difficult places during any disaster and\ncollect high resolution imagery that is required for aforementioned tasks of\ncomputer vision. To address these issues we present a high resolution UAV\nimagery, FloodNet, captured after the hurricane Harvey. This dataset\ndemonstrates the post flooded damages of the affected areas. The images are\nlabeled pixel-wise for semantic segmentation task and questions are produced\nfor the task of visual question answering. FloodNet poses several challenges\nincluding detection of flooded roads and buildings and distinguishing between\nnatural water and flooded water. With the advancement of deep learning\nalgorithms, we can analyze the impact of any disaster which can make a precise\nunderstanding of the affected areas. In this paper, we compare and contrast the\nperformances of baseline methods for image classification, semantic\nsegmentation, and visual question answering on our dataset.\n