Flood10k Dataset: An Image Dataset for Flood Classification in Disaster Response

Flood10K contains 10 419 images (5 093 flooded, 5 326 normal) curated from multiple public sources for training and evaluating binary classification models to identify flood-related imagery in disaster contexts.

Flooded images were aggregated from several publicly available online datasets, including the CrisisMMD datasets, the Twitter Flood Dataset, and the Flooding Image Dataset (FloodIMG). Normal images were sourced through direct downloads from Twitter (1358 images) and extensive Google searches (3968 images).

This dataset provides a robust resource for researchers developing systems to filter and categorize visual content from social media during disaster events, enhancing situational awareness for emergency responders. It supports the development of models capable of distinguishing actionable flood imagery from irrelevant visual noise.