SCRWKV: Ultra-Compact Structure-Calibrated Vision-RWKV for Topological Crack Segmentation

Researchers introduced SCRWKV, a 1.22M-parameter crack-segmentation model built on a structure-calibrated Vision-RWKV backbone.

The network combines multi-scale texture modeling, topology capture, noise suppression, and cross-scale feature fusion for better crack detection.

On the TUT benchmark, it achieved 0.8428 F1 and 0.8512 mIoU, showing strong accuracy despite its tiny size.

The result suggests high-performance crack inspection can be deployed with faster, lighter models for real-world infrastructure monitoring.