List your company
ResearchFormerly WizWand

Image Classification on CIFAR-100-LT Imbalance Ratio 100 (test)

89.1AccuracyLPT
16.986435.708254.4373.1518Dec 13, 2020May 30, 2021Nov 14, 2021May 1, 2022Oct 16, 2022Apr 2, 2023Sep 18, 2023
Updated 8mo ago

Evaluation results

MethodLinks
LPTBackbone=ViT-B/16, Learnable Params.=1.01M, Epochs=40+40, Training Protocol=Fine-tuning pre-trained model from ImageNet-21K2023.0989.1—
LIFTBackbone=ViT-B/16, Learnable Params.=0.10M, Epochs=10, Training Protocol=Fine-tuning pre-trained model from ImageNet-21K2023.0989.1—
LIFT w/ TTEBackbone=ViT-B/16, Learnable Params.=0.10M, Epochs=10, Training Protocol=Fine-tuning pre-trained model2023.0981.7—
LIFTBackbone=ViT-B/16, Learnable Params.=0.10M, Epochs=10, Training Protocol=Fine-tuning pre-trained model2023.0980.2—
BALLADBackbone=ViT-B/16, Learnable Params.=149.62M, Epochs=50+10, Training Protocol=Fine-tuning pre-trained model2023.0977.8—
LiVTBackbone=ViT-B/16, Learnable Params.=85.80M, Epochs=100, Training Protocol=Fine-tuning pre-trained model2023.0958.2—
Meta-SuperDisco2023.0353.8—
Alshammari et al.Venue=CVPR 222023.0353.3—
Cui et al.Venue=ICCV 212023.0352—
PaCoBackbone=ResNet-32, Learnable Params.=0.46M, Epochs=400, Training Protocol=Training from scratch2023.0952—
Zhu et al.Venue=CVPR 222023.0351.9—
BCLBackbone=ResNet-32, Learnable Params.=0.46M, Epochs=200, Training Protocol=Training from scratch2023.0951.9—
BSBackbone=ResNet-32, Learnable Params.=0.46M, Epochs=400, Training Protocol=Training from scratch2023.0950.8—
Wang et al.Venue=ICLR 212023.0348—
Samuel et al.Venue=ICCV 212023.0347.3—
MiSLASBackbone=ResNet-32, Learnable Params.=0.46M, Epochs=200+10, Training Protocol=Training from scratch2023.0947—
Zhong et al.Venue=CVPR 212023.0346.8—
CBD + TailCalibXType=Ours2021.1146.59—
Li et al.Venue=CVPR 212023.0346—
ABSGD (CB-CE)Backbone=ResNet32, Loss Function=CB-CE2020.1245.89—
ABSGD (CB-LDAM)Backbone=ResNet32, Loss Function=CB-LDAM2020.1245.65—
ABSGD (CB-CE)Backbone=ResNet32, Loss Function=CB-CE2020.1245.54—
ABSGD (CB-Focal)Backbone=ResNet32, Loss Function=CB-Focal2020.1245.41—
DiVEBackbone=ResNet-32, Learnable Params.=0.46M, Epochs=200, Training Protocol=Training from scratch2023.0945.4—
SGD (CB-LDAM)Backbone=ResNet32, Loss Function=CB-LDAM2020.1245.36—
CBDType=Distillation, re-implementation=true2021.1144.83—
ABSGD (CB-LDAM)Backbone=ResNet32, Loss Function=CB-LDAM2020.1244.71—
META (CB-Focal)Backbone=ResNet32, Loss Function=CB-Focal2020.1244.7—
CosineCE + TailCalibXType=Ours2021.1144.44—
ABSGD (CB-Focal)Backbone=ResNet32, Loss Function=CB-Focal2020.1244.11—
GBMType=Loss or Re-weighting2021.1144.1—
META (CB-LDAM)Backbone=ResNet32, Loss Function=CB-LDAM2020.1244.08—
Logit adjustmentType=Loss or Re-weighting2021.1143.89—
META (CB-CE)Backbone=ResNet32, Loss Function=CB-CE2020.1243.35—
CosineCE + TailCalibType=Ours2021.1143.03—
M2MType=Generation2021.1142.9—
MODALSType=Generation, re-implementation=true2021.1142.813—
CRTType=Decouple, re-implementation=true2021.1142.63—
BBNBackbone=ResNet-32, Learnable Params.=0.46M, Epochs=200, Training Protocol=Training from scratch2023.0942.6—
BBNType=Decouple2021.1142.56—
LFMEType=Distillation2021.1142.3—
SGD (CB-LDAM)Backbone=ResNet32, Loss Function=CB-LDAM2020.1242.04—
LDAM-DRWType=Loss or Re-weighting2021.1142.04—
Park et al.Venue=ICCV 212023.0342—
LDAMBackbone=ResNet-32, Learnable Params.=0.46M, Epochs=200, Training Protocol=Training from scratch2023.0942—
Meta Weight NetType=Loss or Re-weighting2021.1141.61—
CosineCEType=Baseline, re-implementation=true2021.1141.44—
CEType=Baseline, re-implementation=true2021.1139.89—
SGD (LDAM)Backbone=ResNet32, Loss Function=LDAM2020.1239.6—
Class-Balanced FocalType=Loss or Re-weighting2021.1139.6—
SGD (LDAM)Backbone=ResNet32, Loss Function=LDAM2020.1239.58—
mixupType=Generation2021.1139.54—
Domain AdaptationType=Loss or Re-weighting2021.1139.31—
L2RWType=Loss or Re-weighting2021.1138.9—
ResamplingBackbone=ResNet32, Loss Function=CE2020.1238.87—
ResamplingBackbone=ResNet32, Loss Function=CE2020.1238.84—
SGD (CB-CE)Backbone=ResNet32, Loss Function=CB-CE2020.1238.7—
Focal LossType=Loss or Re-weighting2021.1138.41—
Manifold mixupType=Generation2021.1138.25—
SGD (CB-Focal)Backbone=ResNet32, Loss Function=CB-Focal2020.1236.02—
SGD (CB-CE)Backbone=ResNet32, Loss Function=CB-CE2020.1221.31—
SGD (CB-Focal)Backbone=ResNet32, Loss Function=CB-Focal2020.1219.76—
CB-CEBackbone=ResNet-32, Imbalance Ratio=1002021.11—38.6
CB-FocalBackbone=ResNet-32, Imbalance Ratio=1002021.11—39.6
CEBackbone=ResNet-32, Reproduced=true2021.04—38.32
CEBackbone=ResNet-32, Imbalance Ratio=1002021.11—38.3
CE-DRSBackbone=ResNet-32, Imbalance Ratio=1002021.11—40.4
CE-DRWBackbone=ResNet-32, Imbalance Ratio=1002021.11—40.5
cRTBackbone=ResNet-32, Reproduced=true2021.04—41.24
DRO-LTBackbone=ResNet-32, Radius Config=ε = 0 (ERM)2021.04—43.92
DRO-LTBackbone=ResNet-32, Radius Config=Shared ε2021.04—45.66
DRO-LTBackbone=ResNet-32, Radius Config=ε/√n2021.04—46.92
DRO-LTBackbone=ResNet-32, Radius Config=Learned ε2021.04—47.31
FocalBackbone=ResNet-32, Imbalance Ratio=1002021.11—38.4
Focal LossBackbone=ResNet-322021.04—38.41
KCL†Backbone=ResNet-32, Imbalance Ratio=1002021.11—42.8
LDAMBackbone=ResNet-32, Imbalance Ratio=1002021.11—39.6
LDAM LossBackbone=ResNet-322021.04—39.6
LDAM-DRWBackbone=ResNet-32, Imbalance Ratio=1002021.11—42
M2m-ERMBackbone=ResNet-32, Imbalance Ratio=1002021.11—42.9
M2m-LDAMBackbone=ResNet-32, Imbalance Ratio=1002021.11—43.5
ReSampleBackbone=ResNet-322021.04—33.44
ReWeightBackbone=ResNet-322021.04—33.99
smDRAGONBackbone=ResNet-322021.04—43.55
SSLBackbone=ResNet-32, Reproduced=true2021.04—37.51
SSPBackbone=ResNet-322021.04—43.43
TSCBackbone=ResNet-32, Imbalance Ratio=1002021.11—43.8
τ-normBackbone=ResNet-32, Reproduced=true2021.04—41.11