Adversarial Robustness on ImageNet sr=90% (val)
83.29Clean AccuracyAT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ATBackbone=ResNet-101, Sparsity rate=90%, Params=44.6M2025.04 | 83.29 | 62.23 | 51.34 | 58.65 | 47.29 | 1 | |
| EEDBackbone=ResNet-101, Sparsity rate=90%, Params=4.5M2025.04 | 74.1 | 64.66 | 52.42 | 60.92 | 51.83 | 2.79 | |
| HARPBackbone=ResNet-101, Sparsity rate=90%, Params=4.5M2025.04 | 73.44 | 62.78 | 50.73 | 58.86 | 48.16 | 2.63 | |
| TwinRepBackbone=ResNet-101, Sparsity rate=90%, Params=4.5M2025.04 | 72.72 | 62.14 | 49.46 | 56.47 | 47.77 | 2.8 | |
| Flying BirdBackbone=ResNet-101, Sparsity rate=90%, Params=4.5M2025.04 | 72.56 | 61.32 | 48.94 | 57.73 | 48.24 | 2.61 | |
| ATBackbone=ResNet-50, Sparsity rate=90%, Params=25.6M2025.04 | 60.25 | 32.38 | 28.79 | 30.67 | 34.54 | 1 | |
| EEDBackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 58.41 | 30.54 | 26.89 | 27.15 | 29.92 | 2.84 | |
| HARPBackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 55.21 | 27.1 | 22.57 | 24.62 | 25.57 | 2.88 | |
| TwinRepBackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 52.46 | 26.73 | 24.75 | 24.37 | 24.68 | 2.67 | |
| HYDRABackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 49.44 | 23.75 | 19.88 | 21.6 | 23.14 | 2.73 | |
| Flying BirdBackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 48.39 | 23.05 | 18.14 | 19.33 | 17.65 | 2.56 | |
| R-ADMMBackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 35.26 | 14.35 | 11.01 | 12.35 | 13.53 | 2.41 | |
| MADBackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 34.62 | 14.67 | 11.24 | 12.42 | 13.47 | 2.6 | |
| RSTBackbone=ResNet-50, Sparsity rate=90%, Params=2.6M2025.04 | 27.09 | 12.23 | 10.09 | 11.22 | 12.34 | 2.96 |