DeltaSeg: Tiered Attention and Deep Delta Learning for Multi-Class Structural Defect Segmentation
Enrique Hernandez Noguera, Md Meftahul Ferdaus, Elias Ioup, Mahdi Abdelguerfi, [2pt] University of New Orleans, New Orleans, LA, USA, [2pt] Corresponding author · 2026
Read DeltaSeg: Tiered Attention and Deep Delta Learning for Multi-Class Structural Defect Segmentation by Enrique Hernandez Noguera, Md Meftahul Ferdaus, Elias Ioup, Mahdi Abdelguerfi, [2pt] University of New Orleans, New Orleans, LA, USA, [2pt] Corresponding author (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-18745.txt.
Description
Hernandez and three colleagues present DeltaSeg, a U-shaped encoder-decoder net for finding defects in inspection images. Attention is tiered: channel attention in the encoder, coordinate attention at the bottleneck and decoder, and a deep delta attention step in the skip links. Depthwise separable convolutions with dilated stages keep detail, while multi-scale pooling at the bottleneck takes in context. On two defect data sets with 7 and 9 classes, DeltaSeg beats 12 rival nets. The goal is defect outlines even when damage types vary and some classes are rare.
Catalog record
| CREATOR | Enrique Hernandez Noguera, Md Meftahul Ferdaus, Elias Ioup, Mahdi Abdelguerfi, [2pt] University of New Orleans, New Orleans, LA, USA, [2pt] Corresponding author |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36822 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.18745