Rapid and accurate assessment of flood extent is important for effective disaster response, mitigation planning, and resource allocation. Traditional flood mapping methods encounter challenges in scalability and transferability. However, the emergence of deep learning, particularly convolutional neural networks (CNNs), revolutionizes flood mapping by autonomously learning intricate spatial patterns and semantic features directly from raw data. DeepFlood is introduced to address the essential requirement for high-quality training datasets. This is a novel dataset comprising high-resolution manned and unmanned aerial imagery and Synthetic Aperture Radar (SAR) imagery, enriched with detailed labels including inundated vegetation, one of the most challenging areas for flood mapping. DeepFlood enables multi-modal flood mapping approaches and mitigates limitations in existing datasets by providing comprehensive annotations and diverse landscape coverage. We evaluate several semantic segmentation architectures on DeepFlood, demonstrating its usability and efficacy in post-disaster flood mapping scenarios.