List your company
ResearchFormerly WizWand

Long-tailed Image Classification on CIFAR-100-LT Imbalance ratio 50

57.75Top-1 AccPrototype Classifier
43.29447.04750.854.553Jul 26, 2021Nov 13, 2021Mar 3, 2022Jun 22, 2022Oct 10, 2022Jan 28, 2023May 19, 2023
Updated 8mo ago

Evaluation results

MethodLinks
Prototype Classifier2023.0257.75
WD + WD + Maxweight regularization=WD + WD + Max2023.0257.71
DRO-LT2023.0257.57
PaCOself-supervised pre-training=true, aggressive data augmentation=true2023.0256
PaCoTraining setup=RandAugment [12] in 400 epochs2021.0756
Balanced SoftmaxTraining setup=RandAugment [12] in 400 epochs2021.0754.2
NCM2023.0253.62
GCLBackbone Net=ResNet-322023.0553.55
WDweight decay=tuned baseline2023.0252.71
MisLASBackbone Net=ResNet-322023.0552.62
MISLAS2021.0752.3
Contrastive learningBackbone Net=ResNet-322023.0551.87
CE loss + mixup + cRTBackbone Net=ResNet-322023.0550.86
De-confound-TDEBackbone Net=ResNet-322023.0550.31
Causal Norm2021.0750.3
ResLT2021.0750
LDAM-DRWBackbone Net=ResNet-322023.0547.62
LogitAdjust2023.0247.03
BBNBackbone Net=ResNet-322023.0547.02
BBN2021.0747
cRT2023.0246.83
LDAM+DRW2021.0746.6
CE loss + mixupBackbone Net=ResNet-322023.0545.16
LDAM Loss2023.0244.97
Focal Loss2023.0244.32
Focal Loss2021.0744.3
CE lossBackbone Net=ResNet-322023.0543.9
CE2023.0243.85