Table 2 Experimental results on the CLTP-DD.
From: Dual-branch information extraction and local attention anchor-free network for defect detection
Network | Backbone | Params (M) | GFLOPs | mAP (%) | FPS (img/s) |
|---|---|---|---|---|---|
Faster R-CNN | ResNet50 | 164.96 | 379.69 | 78.7 | 5.4 |
SSD | VGG16 | 24.53 | 87.86 | 93.8 | 28.3 |
YOLO V3 | Darknet53 | 61.53 | 193.87 | 93.6 | 12.8 |
YOLO V5-L | CSPDarknet53 | 46.14 | 53.98 | 94.2 | 37.6 |
RetinaNet | Swin-T | 36.84 | 210.29 | 80.5 | 17.2 |
Deformable DETR | ResNet50 | 39.82 | 195.23 | 95.1 | 10.1 |
YOLOX-S | CSPDarknet53 | 8.94 | 33.3 | 91.3 | 16.9 |
DINO | ResNet50 | 47.54 | 197.00 | 95.3 | 25.1 |
Mask R-CNN | Swin-T | 47.38 | 261.81 | 95.6 | 9.7 |
SwinTD | Swin-T | 47.38 | 262.9 | 96.4 | 9.6 |
FCOS | ResNet50 | 32.13 | 125.95 | 93.8 | 22.0 |
DLA-FCOS(Ours) | DFENet | 41.01 | 186.37 | 96.8 | 20.7 |
