FloodNet: A High Resolution Aerial Imagery Dataset for Post Flood Scene Understanding

Image | Hurricane | Response

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.

  • ML task type: Image classification, semantic segmentation, visual question answering
  • Data Source: Earth Observation Data and GeoSpatial Imagery (UAV)
  • Size: ∼11, 000 Question-image pairs for VQA; 3,200 Images
  • Timespan: August 30 - September 4, 2017
  • Geographical Coverage: Ford Bend County in Texas and other directly impacted areas from Hurricane Harvey
  • Image Classification
  • Evaluated on: InceptionNetv3, ResNet50, Xception
  • Metrics used: accuracy
  • Results as reported in original paper: Training Accuracy: 99.84% (Xception), Test Accuracy: 93.69% (ResNet50)
  • Semantic Segementation
  • Evaluated on: PSPNet, ENet,, DeepLabv3+
  • Metrics used: Mean Intersection over Union
  • Results as reported in original paper: mIoU: 79.69 (PSPNet)
  • Visual Question Answering
  • Evaluated on: Concatenation of Features, Element-wise Multiplication of Features, SAN, MFB with Co-Attention
  • Metrics used: Overall Accuracy
  • Results as reported in original paper: Overall Accuracy: 0.73 (MFB with Co-Attention)

Maryam Rahnemoonfar, Tashnim Chowdhury, Argho Sarkar, Debvrat Varshney, Masoud Yari, and Robin Roberson Murphy. FloodNet: A high resolution aerial imagery dataset for post flood scene understanding. IEEE Access, 9:89644– 89654, 2021.