Table 2 Detailed parameter configurations of the models used for SHL
From: Mitigating data bias and ensuring reliable evaluation of AI models with shortcut hull learning
Model | Layer | Size | Params (M) | FLOPs (G) |
|---|---|---|---|---|
ResNet-5026 | layer3.0 | 14 × 14 | 23.51 | 4.12 |
ViT-B/1641 | layers.8 | 14 × 14 | 88.17 | 16.86 |
RepVGG-A257 | stage_3.0 | 14 × 14 | 25.50 | 5.12 |
Swin-T58 | stages.2.blocks.0 | 14 × 14 | 27.52 | 4.36 |
PViG-S59 | stages.2.0 | 14 × 14 | 29.02 | 4.57 |
ResNeXt-5068 | layer3.0 | 14 × 14 | 25.03 | 4.27 |
Inception-V363 | Mixed_6a | 12 × 12 | 23.83 | 5.75 |
ConvMixer-1024/1069 | stages.8 | 16 × 16 | 24.38 | 5.55 |
EfficientNet-B470 | layers.4.0 | 14 × 14 | 19.34 | 4.66 |
RegNetX-4.0GF71 | layer3.0 | 14 × 14 | 22.12 | 4.00 |
SE-ResNet-5072 | layer3.0 | 14 × 14 | 28.09 | 4.13 |
