Multi-class segmentation under severe class imbalance: A case study in roof damage assessment
This work addresses a domain-specific problem of roof damage assessment from overhead imagery, focusing on improving segmentation under class imbalance, which is incremental as it builds on existing methods.
The paper tackled the problem of severe class imbalance in multi-class segmentation for roof damage assessment by proposing four techniques, including oversampling minority classes and network architectural improvements, which boosted the macro-averaged F1-score by 39.9 percentage points.
The task of roof damage classification and segmentation from overhead imagery presents unique challenges. In this work we choose to address the challenge posed due to strong class imbalance. We propose four distinct techniques that aim at mitigating this problem. Through a new scheme that feeds the data to the network by oversampling the minority classes, and three other network architectural improvements, we manage to boost the macro-averaged F1-score of a model by 39.9 percentage points, thus achieving improved segmentation performance, especially on the minority classes.