Image Classification on ImageNet 9 (val)
95.7Top-1 AccuracyCaaM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CaaMspurious attribute label=false2022.06 | 95.7 | — | 95.2 | |
| SSL+ERMspurious attribute label=false2022.06 | 94.18 | — | 93.18 | |
| LWBCspurious attribute label=false2022.06 | 94.03 | — | 93.04 | |
| ReBiasspurious attribute label=false2022.06 | 91.9 | — | 90.5 | |
| ERMspurious attribute label=false2022.06 | 90.8 | — | 88.8 | |
| RUBispurious attribute label=true2022.06 | 90.5 | — | 88.6 | |
| StylisedINspurious attribute label=true2022.06 | 88.4 | — | 86.6 | |
| LfFspurious attribute label=false2022.06 | 86 | — | 85 | |
| AA-ResNet-152Architecture=ResNet-152, GFlops=23.8, Params=61.6M2019.04 | 79.1 | 94.6 | — | |
| SE-ResNet-152Architecture=ResNet-152, GFlops=23.1, Params=66.8M2019.04 | 78.9 | 94.5 | — | |
| AA-ResNet-101Architecture=ResNet-101, GFlops=16.1, Params=45.4M2019.04 | 78.7 | 94.4 | — | |
| SE-ResNet-101Architecture=ResNet-101, GFlops=15.6, Params=49.3M2019.04 | 78.4 | 94.2 | — | |
| ResNet-152Architecture=ResNet-152, GFlops=23, Params=60.2M2019.04 | 78.4 | 94.2 | — | |
| ResNet-101Architecture=ResNet-101, GFlops=15.6, Params=44.5M2019.04 | 77.9 | 94 | — | |
| AA-ResNet-50Architecture=ResNet-50, GFlops=8.3, Params=25.8M2019.04 | 77.7 | 93.8 | — | |
| SE-ResNet-50Architecture=ResNet-50, GFlops=8.2, Params=28.1M2019.04 | 77.5 | 93.7 | — | |
| ResNet-50Architecture=ResNet-50, GFlops=8.2, Params=25.6M2019.04 | 76.4 | 93.1 | — | |
| AA-ResNet-34Architecture=ResNet-34, GFlops=7.1, Params=20.7M2019.04 | 74.7 | 92 | — | |
| SE-ResNet-34Architecture=ResNet-34, GFlops=7.4, Params=22.0M2019.04 | 74.3 | 91.8 | — | |
| ResNet-34Architecture=ResNet-34, GFlops=7.4, Params=21.8M2019.04 | 73.6 | 91.5 | — | |
| MobileNetV1-0.25 with 5 x 5 filtersWidth Multiplier=0.25, Filter Size=5x52021.04 | 69.44 | 89.9 | — | |
| MobileNetV1-0.25 with dilated filters (B)Width Multiplier=0.25, Filter Type=dilated, Layers=14 to 18, Kernel size=3x3, Dilation rates=2, 3, 2, 3, 22021.04 | 68.98 | 88.85 | — | |
| MobileNetV1-0.25 with dilated filters (A)Width Multiplier=0.25, Filter Type=dilated, Layers=14 to 18, Kernel size=3x3, Dilation rates=2, 2, 2, 2, 22021.04 | 68.52 | 88.61 | — | |
| MobileNetV1-0.25Width Multiplier=0.252021.04 | 68.39 | 88.35 | — | |
| Biased (BagNet18)spurious attribute label=false2022.06 | 67.7 | — | 65.9 | |
| LearnedMixinspurious attribute label=true2022.06 | 64.1 | — | 62.7 |