SEMIR: Topology-Preserving Graph Minors for Thin-Structure Segmentation

Researchers propose SEMIR, a framework that replaces the pixel lattice with a parameterized graph minor to preserve thin-structure connectivity in segmentation tasks. The method achieves full-resolution inference without patching, matching or exceeding domain-specific baselines on power line, crack, and lane marking datasets while reducing mask fragmentation by at least 4.6x.

arXiv:2606.24935v1 Announce Type: new Abstract: Thin-structure segmentation--power lines, cracks, lane markings at 1-3 pixel width--requires preserving connectivity that standard representations preclude: patching severs continuous structures and conventional superpixels merge thin targets into background before classification. Topology-aware losses penalize connectivity breaks at the objective level but cannot recover what the representation has already destroyed. We propose SEMIR, a framework that replaces the pixel lattice with a parameterized graph minor whose contraction map preserves thin-structure connectivity under the contraction criterion. The minor collapses millions of pixels into tens or hundreds of boundary-aligned supernodes, enabling full-resolution inference without patching at scales demonstrated up to 21 MP in this paper; a lightweight GNN classifies the reduced graph and an exact map lifts predictions to pixel resolution. One pipeline--identical architecture, features, loss, and GNN hyperparameters across all dataset--matches or exceeds domain-specific baselines on TTPLA power lines , CrackSeg9k pavement cracks , and SkyScapes Lane aerial markings on Dice, IoU, and Boundary F1 while reducing mask fragmentation by at least 4.6x relative to SLIC at matched inference.