Physics-aware Masked Diffusion-based Flood Simulation for Urban Fisheye Disaster Detection
PhysFlood synthesizes realistic flood images from a single fisheye capture using Diffusion Models and physical constraints. It uses a Masker to estimate plausible water level masks based on ground segmentation and height estimation, allowing users to control the flood depth precisely via prompts. This method successfully generates physically consistent floods matching specified water levels, outperforming standard methods that ignore these physical parameters.