Image Classification on CIFAR100-LT (test) with IR-based Accuracy
62.1Top-1 Acc (Avg)FixMatch w/ DECON
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FixMatch w/ DECONImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 62.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ ACRImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 61.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ CDMADImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 61 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ FARADImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 60.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ DARPImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 58.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ CReST+Imbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 57.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Dynamic Loss + Balanced-SoftmaxBackbone=ResNet322022.11 | 57.36 | 50.54 | 54.6 | 64.18 | 60.13 | — | — | — | — | — | — | — | — | |
| Dynamic LossBackbone=ResNet322022.11 | 57.11 | 50.14 | 54.51 | 63.99 | 59.79 | — | — | — | — | — | — | — | — | |
| Balanced-SoftmaxBackbone=ResNet322022.11 | 57.08 | 50.47 | 54.36 | 64 | 59.48 | — | — | — | — | — | — | — | — | |
| Dynamic Loss + Logit AdjustmentBackbone=ResNet322022.11 | 57.01 | 49.65 | 54.14 | 64.15 | 60.08 | — | — | — | — | — | — | — | — | |
| FixMatch w/ DECONImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 56.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatchImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 56.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ ACRImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 55.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ FARADImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 55.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Logit AdjustmentBackbone=ResNet322022.11 | 55.54 | 48.36 | 52.15 | 62.83 | 58.81 | — | — | — | — | — | — | — | — | |
| MiSLASBackbone=ResNet322022.11 | 55.44 | 47 | 52.3 | 63.2 | 59.25 | — | — | — | — | — | — | — | — | |
| CMOBackbone=ResNet322022.11 | 55.11 | 46.6 | 51.4 | 62.3 | 60.12 | — | — | — | — | — | — | — | — | |
| FixMatch w/ CDMADImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 54.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ DASOImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 52.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FaMUSBackbone=ResNet322022.11 | 52.73 | 46.03 | 49.93 | 59 | 55.95 | — | — | — | — | — | — | — | — | |
| FixMatch w/ DARPImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 52.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ CReST+Imbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 52.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ DECONImbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 52 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LDAM-DRWBackbone=ResNet322022.11 | 51.36 | 59.59 | 48.22 | 44.7 | 52.93 | — | — | — | — | — | — | — | — | |
| FixMatch w/ ACRImbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 51.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatchImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 50.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| WDBackbone=ResNet322022.11 | 50.26 | 40.79 | 45.89 | 61.6 | 52.75 | — | — | — | — | — | — | — | — | |
| FixMatch w/ DASOImbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 49.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ DARPImbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 49.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ DASOImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 49.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CB FocalBackbone=ResNet322022.11 | 48.88 | 39.6 | 45.32 | 57.99 | 52.59 | — | — | — | — | — | — | — | — | |
| Supervised w/ LAImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 48.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Focal LossBackbone=ResNet322022.11 | 47.62 | 38.41 | 44.32 | 55.78 | 51.95 | — | — | — | — | — | — | — | — | |
| Cross entropyBackbone=ResNet322022.11 | 47.26 | 38.32 | 43.85 | 55.71 | 51.14 | — | — | — | — | — | — | — | — | |
| SupervisedImbalance ratio (γ)=10, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 46.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| IGLU-Approxsigma=0.5, Backbone=ResNet-202026.03 | 45.82 | — | — | — | — | — | — | — | — | — | 3.074 | — | — | |
| FixMatch w/ DECONImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 45.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatchImbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 45.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| IGLUsigma=0.5, Backbone=ResNet-202026.03 | 44.97 | — | — | — | — | — | — | — | — | — | 3.309 | — | — | |
| FixMatch w/ ACRImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 44.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| GELUBackbone=ResNet-202026.03 | 44.64 | — | — | — | — | — | — | — | — | — | 3.751 | — | — | |
| MishBackbone=ResNet-202026.03 | 44.56 | — | — | — | — | — | — | — | — | — | 3.648 | — | — | |
| FixMatch w/ CReST+Imbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 44.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| IGLU-Approxsigma=1, Backbone=ResNet-202026.03 | 44.49 | — | — | — | — | — | — | — | — | — | 3.458 | — | — | |
| SiLUBackbone=ResNet-202026.03 | 44.15 | — | — | — | — | — | — | — | — | — | 3.737 | — | — | |
| Supervised w/ LAImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 44.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ReLUBackbone=ResNet-202026.03 | 43.99 | — | — | — | — | — | — | — | — | — | 3.57 | — | — | |
| IGLUsigma=1, Backbone=ResNet-202026.03 | 43.95 | — | — | — | — | — | — | — | — | — | 3.514 | — | — | |
| IGLU-Approxsigma=5, Backbone=ResNet-202026.03 | 43.74 | — | — | — | — | — | — | — | — | — | 3.558 | — | — | |
| HardswishBackbone=ResNet-202026.03 | 43.65 | — | — | — | — | — | — | — | — | — | 3.761 | — | — | |
| IGLU-Approxsigma=10, Backbone=ResNet-202026.03 | 43.64 | — | — | — | — | — | — | — | — | — | 3.622 | — | — | |
| IGLUsigma=5, Backbone=ResNet-202026.03 | 43.63 | — | — | — | — | — | — | — | — | — | 3.675 | — | — | |
| FixMatch w/ DASOImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 43.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| IGLUsigma=10, Backbone=ResNet-202026.03 | 43.49 | — | — | — | — | — | — | — | — | — | 3.57 | — | — | |
| FixMatch w/ DARPImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 43.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatch w/ SimProImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 43.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SupervisedImbalance ratio (γ)=20, Labeled samples (N1)=150, Unlabeled samples (M1)=3002024.06 | 41.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| IGLU-Approxsigma=0.1, Backbone=ResNet-202026.03 | 40.75 | — | — | — | — | — | — | — | — | — | 2.39 | — | — | |
| FixMatch w/ CReST+Imbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 40.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| FixMatchImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 40 | — | — | — | — | — | — | — | — | — | — | — | — | |
| IGLUsigma=0.1, Backbone=ResNet-202026.03 | 38.6 | — | — | — | — | — | — | — | — | — | 2.456 | — | — | |
| Supervised w/ LAImbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 30.2 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SupervisedImbalance ratio (γ)=10, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 29.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Supervised w/ LAImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 26.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SupervisedImbalance ratio (γ)=20, Labeled samples (N1)=50, Unlabeled samples (M1)=4002024.06 | 25.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| AT-BSLImbalance Ratio (IR)=202025.03 | — | — | — | — | — | 48.27 | 24.98 | 20.23 | 20.05 | 18.12 | — | — | — | |
| BBNBackbone=ResNet-322021.10 | — | 42.6 | 47 | 59.1 | — | — | — | — | — | — | — | — | — | |
| BBNBackbone=ResNet-32, Method Category=rebalance classifier2023.05 | — | 42.56 | 47.02 | 59.12 | — | — | — | — | — | — | — | — | — | |
| BCLBackbone=ResNet-32, Method Category=self-supervised pretraining2023.05 | — | 51.93 | 56.59 | 64.87 | — | — | — | — | — | — | — | — | — | |
| CB-FocalBackbone=ResNet-322021.10 | — | 39.6 | 45.2 | 58 | — | — | — | — | — | — | — | — | — | |
| CB-FocalBackbone=ResNet-32, Method Category=rebalance classifier2023.05 | — | 39.6 | 45.2 | 58 | — | — | — | — | — | — | — | — | — | |
| CEBackbone=ResNet-322021.10 | — | 38.3 | 43.9 | 55.7 | — | — | — | — | — | — | — | — | — | |
| CEBackbone=ResNet-322023.05 | — | 38.3 | 43.9 | 55.7 | — | — | — | — | — | — | — | — | — | |
| CE-DRSBackbone=ResNet-322021.10 | — | 41.6 | 45.5 | 58.1 | — | — | — | — | — | — | — | — | — | |
| CE-DRWBackbone=ResNet-322021.10 | — | 41.5 | 45.3 | 58.1 | — | — | — | — | — | — | — | — | — | |
| CMOBackbone=ResNet-32, Method Category=augmentation2023.05 | — | 47.2 | 51.7 | 58.4 | — | — | — | — | — | — | — | — | — | |
| cRTBackbone=ResNet-322021.10 | — | 42.3 | 46.8 | 58.1 | — | — | — | — | — | — | — | — | — | |
| De-confound-TDEBackbone=ResNet-322021.10 | — | 44.1 | 50.3 | 59.6 | — | — | — | — | — | — | — | — | — | |
| FixMatchγl=20, γu=20, N1=50, M1=4002024.06 | — | — | — | — | — | — | — | — | — | — | — | 39.7 | 38.2 | |
| FixMatch w/ ACRγl=20, γu=20, N1=50, M1=4002024.06 | — | — | — | — | — | — | — | — | — | — | — | 44.6 | 42.3 | |
| FixMatch w/ CReST+γl=20, γu=20, N1=50, M1=4002024.06 | — | — | — | — | — | — | — | — | — | — | — | 36.9 | 35.1 | |
| FixMatch w/ DASOγl=20, γu=20, N1=50, M1=4002024.06 | — | — | — | — | — | — | — | — | — | — | — | 43.1 | 43.8 | |
| FixMatch w/ DECONγl=20, γu=20, N1=50, M1=4002024.06 | — | — | — | — | — | — | — | — | — | — | — | 45.7 | 44.9 | |
| FixMatch w/ SimProγl=20, γu=20, N1=50, M1=4002024.06 | — | — | — | — | — | — | — | — | — | — | — | 43.6 | 44.8 | |
| Focal LossBackbone=ResNet-322021.10 | — | 38.4 | 44.3 | 55.8 | — | — | — | — | — | — | — | — | — | |
| GLMCBackbone=ResNet-32, Training Stage=one-stage training2023.05 | — | 55.88 | 61.08 | 70.74 | — | — | — | — | — | — | — | — | — | |
| GLMC + MaxNormBackbone=ResNet-32, Training Stage=finetune classifier2023.05 | — | 57.11 | 62.32 | 72.33 | — | — | — | — | — | — | — | — | — | |
| ICCLBackbone=ResNet-322021.10 | — | 46.6 | 51.6 | 62.1 | — | — | — | — | — | — | — | — | — | |
| KCLBackbone=ResNet-32, Method Category=self-supervised pretraining2023.05 | — | 42.8 | 46.3 | 57.6 | — | — | — | — | — | — | — | — | — | |
| LDAM-DRWBackbone=ResNet-322021.10 | — | 42 | 46.6 | 58.7 | — | — | — | — | — | — | — | — | — | |
| LogitAjustBackbone=ResNet-32, Method Category=rebalance classifier2023.05 | — | 42.01 | 47.03 | 57.74 | — | — | — | — | — | — | — | — | — | |
| LWSBackbone=ResNet-322021.10 | — | 42.3 | 46.4 | 58.1 | — | — | — | — | — | — | — | — | — | |
| M2mBackbone=ResNet-322021.10 | — | 43.5 | — | 57.6 | — | — | — | — | — | — | — | — | — | |
| Manifold MixupBackbone=ResNet-322021.10 | — | 38.3 | 43.1 | 56.6 | — | — | — | — | — | — | — | — | — | |
| Manifold Mixup (two samplers)Backbone=ResNet-322021.10 | — | 36.8 | 42.1 | 56.5 | — | — | — | — | — | — | — | — | — | |
| MixupBackbone=ResNet-322021.10 | — | 39.5 | 45 | 58 | — | — | — | — | — | — | — | — | — | |
| MixupBackbone=ResNet-32, Method Category=augmentation2023.05 | — | 39.54 | 54.99 | 58.02 | — | — | — | — | — | — | — | — | — | |
| PaCoBackbone=ResNet-32, Method Category=self-supervised pretraining2023.05 | — | 52 | 56 | 64.2 | — | — | — | — | — | — | — | — | — | |
| RIDE (3 experts)Backbone=ResNet-32, Method Category=ensemble classifier2023.05 | — | 48.6 | 51.4 | 59.8 | — | — | — | — | — | — | — | — | — | |
| RIDE (3 experts) + CMOBackbone=ResNet-32, Method Category=ensemble classifier2023.05 | — | 50 | 53 | 60.2 | — | — | — | — | — | — | — | — | — |