Sci-Net: scale-invariant model for buildings segmentation from aerial imagery

Buildings’ segmentation is a fundamental task in the field of earth observation and aerial imagery analysis. Most existing deep learning-based methods literature can be applied to fixed or narrow-range spatial resolution imagery. In practical scenarios, users deal with broad spectrum image resolutions. Thus, given often needs re-sampled match dataset used train learning model, which results degradation performance. To overcome this challenge, we propose, manuscript, scale-invariant neural network (Sci-Net) architecture that segments buildings from wide-range images. Specifically, our approach leverages UNet hierarchical representation dense atrous pyramid pooling extract fine-grained multi-scale representations. Sci-Net significantly outperforms state-of-the-art models on open cities AI building datasets steady improvement margin across different