Table 3 Results of experiments with different lightweight backbones.
From: Advanced lightweight deep learning vision framework for efficient pavement damage identification
Backbone | Params/M | GFLOPs | F1 | mAP | FPS |
|---|---|---|---|---|---|
Baseline | 7.02 | 15.8 | 66.3% | 67.6% | 90 |
MobileNetV3 | 5.21 | 10.3 | 65.0% | 65.4% | 66 |
MobileNetV2 | 4.55 | 10.0 | 62.9% | 66.5% | 69 |
ShuffleNetV2 | 6.4 | 14.1 | 62.9% | 62.5% | 79 |
GhostNet | 3.69 | 8.1 | 67.2% | 68.0% | 85 |
Im-FasterNet | 3.7 | 7.2 | 71.6% | 71.7% | 85 |
