List your company
ResearchFormerly WizWand

Robustness on ImageNet-C

42.5mCECPolyNeXt-L
41.07250.71160.3569.989Oct 28, 2021Aug 1, 2022May 6, 2023Feb 7, 2024Nov 11, 2024Aug 15, 2025May 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
CPolyNeXt-LModel Scale=Large models (~50–90M params), Activation-free=true, Fine-tuned=false2026.0542.5
CAFormer-M36Model Scale=Large models (~50–90M params), Activation-free=false, Fine-tuned=false2026.0542.6
APolyNeXt-BModel Scale=Medium models (~35–50M params), Activation-free=true, Fine-tuned=false2026.0542.7
APolyNeXt-LModel Scale=Large models (~50–90M params), Activation-free=true, Fine-tuned=false2026.0542.9
CPolyNeXt-BModel Scale=Medium models (~35–50M params), Activation-free=true, Fine-tuned=false2026.0544.5
CAFormer-S36Model Scale=Medium models (~35–50M params), Activation-free=false, Fine-tuned=false2026.0544.7
APolyNeXt-SModel Scale=Small models (~25–30M params), Activation-free=true, Fine-tuned=false2026.0545
ConvFormer-M36Model Scale=Large models (~50–90M params), Activation-free=false, Fine-tuned=false2026.0546.5
ConvNeXt-BModel Scale=Large models (~50–90M params), Activation-free=false, Fine-tuned=false2026.0546.8
ConvFormer-S36Model Scale=Medium models (~35–50M params), Activation-free=false, Fine-tuned=false2026.0547.1
CAFormer-S18Model Scale=Small models (~25–30M params), Activation-free=false, Fine-tuned=false2026.0547.4
CPolyNeXt-SModel Scale=Small models (~25–30M params), Activation-free=true, Fine-tuned=false2026.0547.9
MONet-SModel Scale=Medium models (~35–50M params), Activation-free=true, Fine-tuned=false2026.0549.7
ConvNeXt-SModel Scale=Medium models (~35–50M params), Activation-free=false, Fine-tuned=false2026.0551.2
ConvFormer-S18Model Scale=Small models (~25–30M params), Activation-free=false, Fine-tuned=false2026.0551.7
Swin-SModel Scale=Medium models (~35–50M params), Activation-free=false, Fine-tuned=false2026.0552.7
ConvNeXt-TModel Scale=Small models (~25–30M params), Activation-free=false, Fine-tuned=false2026.0553.2
Swin-BModel Scale=Large models (~50–90M params), Activation-free=false, Fine-tuned=false2026.0554.4
Swin-T + ARM#Params=28.4M, FLOPs=4.6G2021.1059.8
Swin-T#Params=28.3M, FLOPs=4.5G2021.1060.7
Swin-TModel Scale=Small models (~25–30M params), Activation-free=false, Fine-tuned=false2026.0562
PIXMIXBackbone=ResNet-50, k=4, beta=42021.1265.8
SINBackbone=ResNet-502021.1270.9
AugMixBackbone=ResNet-502021.1271
MixupBackbone=ResNet-502021.1272.7
AA Res-50#Params=25.6M, FLOPs=4.2G2021.1073.4
AutoAugmentBackbone=ResNet-502021.1273.8
CutoutBackbone=ResNet-502021.1276.9
CutMixBackbone=ResNet-502021.1277.8
BaselineBackbone=ResNet-502021.1278.2