Table 4 Results on the NEU-DET dataset compared with SOTA defect detectors.
From: Dual-branch information extraction and local attention anchor-free network for defect detection
Method | Crazing | Inclusion | Patches | Pitted surface | Rolled in scale | Scratches | mAP (%) |
|---|---|---|---|---|---|---|---|
41.7 | 76.3 | 86.3 | 85.1 | 58.1 | 85.6 | 72.4 | |
55.4 | 75.0 | 93.5 | 88.9 | 62.9 | 84.4 | 76.7 | |
38.9 | 73.7 | 93.5 | 74.8 | 60.7 | 91.4 | 72.2 | |
56.9 | 80.8 | 93.5 | 82.1 | 52.7 | 83.5 | 75.2 | |
60.9 | 82.5 | 94.3 | 95.8 | 67.2 | 74.1 | 79.1 | |
SAPD60 | 60.9 | 82.5 | 93.3 | 87.4 | 42.9 | 97.8 | 73.2 |
MSC-Dnet11 | 42.4 | 84.5 | 94.3 | 91.5 | 71.6 | 92.0 | 79.4 |
DLA-FCOS (Ours) | 56.7 | 89.5 | 91.2 | 83.7 | 65.9 | 83.2 | 78.4 |
