Scale-Invariant Multi-Level Context Aggregation Network for Weakly Supervised Building Extraction

Weakly supervised semantic segmentation (WSSS) methods, utilizing only image-level annotations, are gaining popularity for automated building extraction due to their advantages in eliminating the need costly and time-consuming pixel-level labeling. Class activation maps (CAMs) crucial weakly methods generate pseudo-pixel-level labels training networks segmentation. However, CAMs activate most discriminative regions, leading inaccurate incomplete results. To alleviate this, we propose a scale-invariant multi-level context aggregation network improve quality of terms fineness completeness. The proposed method has integrated two novel modules into Siamese network: (a) self-attentive module that generates attentively aggregates create fine-structured (b) optimization cooperates with mutual learning coarse-to-fine completeness CAMs. results experiments on open datasets demonstrate our achieves new state-of-the-art using labels, producing more complete accurate an IoU 0.6339 WHU dataset 0.5887 Chicago dataset, respectively.