Perturbation Detection on Cityscapes Dcs (val)
98.2Global AuROC (FGSM)Maag et al.
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Maag et al.Optim. on=FGSM, Detector Type=Single-pass detector, Backbone=SegFormer MiT-B02026.06 | 98.2 | 94.9 | 48 | 67.6 | 64.3 | 74.8 | 122 | |
| Xu et al.Detector Type=Multi-pass detector, Backbone=SegFormer MiT-B02026.06 | 97.1 | 98.2 | 78 | 64.2 | 65.4 | 80.7 | 366 | |
| HadamardNet: Quantile (19)Detector Type=Single-pass detector, Backbone=SegFormer MiT-B02026.06 | 95.9 | 96.1 | 75.9 | 68.5 | 64.4 | 80.2 | 124 | |
| EntropyDetector Type=Single-pass detector, Backbone=SegFormer MiT-B02026.06 | 95.7 | 92.2 | 45 | 62.9 | 62.1 | 71.6 | 122 | |
| HadamardNet: Error (17)Detector Type=Single-pass detector, Backbone=SegFormer MiT-B02026.06 | 95.7 | 95.2 | 68.4 | 65 | 60.6 | 77.1 | 123 | |
| Max posteriorDetector Type=Single-pass detector, Backbone=SegFormer MiT-B02026.06 | 95.5 | 91.7 | 45.2 | 62.8 | 61.6 | 71.4 | 122 | |
| HadamardNet: Reg. (18)Detector Type=Single-pass detector, Backbone=SegFormer MiT-B02026.06 | 89.6 | 94.1 | 72.7 | 59.9 | 55.9 | 74.6 | 124 |