Fishdrink / SC-Net

(1) Purpose: A weakly supervised surface defect detection model using image-level labels for simultaneous classification and segmentation. (2) Experiments: run on 4 datasets, including KolektorSDD2(KSDD2), DAGM 1-10, KolektorSDD(KSDD), Severstal Steel, the classification average precision (AP) reaches 96.0%, 100%, 97.1%, 97.7%,respectively. (1)用…
18Updated 11 months ago

Related projects

Alternatives and complementary repositories for SC-Net