| --- |
| license: unknown |
| tags: |
| - anomaly-detection |
| - steel |
| - object-detection |
| - pointer-card |
| pretty_name: NEU-DET (pointer card) |
| --- |
| |
| # NEU-DET — pointer card, not a data mirror |
|
|
| Metadata/citation card only. NEU-DET is hosted on **IEEE DataPort**, which requires an |
| account/subscription to download; no explicit redistribution license was found on the page, so |
| it is not re-hosted here. |
|
|
| ## Access |
| - Official page (account required): https://ieee-dataport.org/documents/neu-det |
| - Related benchmark review: arXiv:2305.13261 |
|
|
| ## Contents |
| 1,800 grayscale images (200x200), 6 defect classes x 300 (crazing, inclusion, patches, pitted |
| surface, rolled-in scale, scratches), **bounding-box** annotated. Bare hot-rolled steel, not |
| painted — useful for bbox-supervised detector warmup (scratch/inclusion classes overlap with |
| paint-defect taxonomy) but not a domain match. |
|
|
| ## Citation |
| Song, K., Yan, Y. "A noise robust method based on completed local binary patterns for hot-rolled |
| steel strip surface defects." Applied Surface Science, 2013. |
|
|
