Revisiting Classical Bagging with Modern Transfer Learning for On-the-fly Disaster Damage Detector
This work addresses the challenge of data-efficient disaster damage detection for emergency responders, though it appears incremental by combining established techniques.
The paper tackles the problem of imbalanced data in automatic post-disaster damage detection from aerial imagery by revisiting classical bagging with modern transfer learning, achieving significant performance improvements over existing methods on the AIST Building Change Detection dataset.
Automatic post-disaster damage detection using aerial imagery is crucial for quick assessment of damage caused by disaster and development of a recovery plan. The main problem preventing us from creating an applicable model in practice is that damaged (positive) examples we are trying to detect are much harder to obtain than undamaged (negative) examples, especially in short time. In this paper, we revisit the classical bootstrap aggregating approach in the context of modern transfer learning for data-efficient disaster damage detection. Unlike previous classical ensemble learning articles, our work points out the effectiveness of simple bagging in deep transfer learning that has been underestimated in the context of imbalanced classification. Benchmark results on the AIST Building Change Detection dataset show that our approach significantly outperforms existing methodologies, including the recently proposed disentanglement learning.