SeMAnD: Self-Supervised Anomaly Detection in Multimodal Geospatial Datasets

Geospatial datasets are diverse, naturally spatiotemporal, and inherently multimodal (composed of two or more distinct signal types or modalities) e.g., satellite/aerial imagery (RGB, multispectral), road network graphs, vector geometry, sensor data (LiDAR-based point cloud, IMU data, GPS-based mobility data), active sensing imagery (SAR, Sonar, Radar).