Damage-TriageFormer: A Foundation-Model Framework for Typology-Based Building Damage Assessment from Mono-Temporal Imagery
Researchers have developed Damage-TriageFormer, a new foundation model designed to assess building damage from single post-disaster images. This model categorizes damage into specific typologies, such as roof or structural damage, rather than a general severity scale. It was trained and evaluated on a new benchmark, DamageTriage-Bench, which includes data from hurricanes and wildfires, achieving a macro F1 score of 0.619 on a test set. The system shows particular strength in identifying undamaged buildings and total structural collapse, aiding in targeted emergency response. AI
IMPACT Enables more precise disaster response and resource allocation using single-image analysis.