Clustering on CIFAR100-20
61.4ACCProPos
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ProPosTraining paradigm=improving representation learning2021.11 | 61.4 | 0.606 | 45.1 | |
| BYOLTraining paradigm=learning general representations2021.11 | 56.9 | 0.559 | 39.3 | |
| TCLTraining paradigm=directly outputting cluster assignments, Image size=224x2242021.11 | 53.1 | 0.529 | 35.7 | |
| PCLTraining paradigm=improving representation learning2021.11 | 52.6 | 0.528 | 36.3 | |
| SCANTraining paradigm=multi-stage methods2021.11 | 50.7 | 0.486 | 33.3 | |
| TCCTraining paradigm=directly outputting cluster assignments2021.11 | 49.1 | 0.479 | 31.2 | |
| SimSiamTraining paradigm=learning general representations2021.11 | 48.5 | 0.522 | 32.7 | |
| NMMTraining paradigm=multi-stage methods2021.11 | 47.7 | 0.484 | 31.6 | |
| GCCTraining paradigm=directly outputting cluster assignments2021.11 | 47.2 | 0.472 | 30.5 | |
| MiCETraining paradigm=directly outputting cluster assignments2021.11 | 44 | 0.436 | 28 | |
| CCTraining paradigm=directly outputting cluster assignments, Image size=224x2242021.11 | 42.9 | 0.431 | 26.6 | |
| IDFDTraining paradigm=improving representation learning2021.11 | 42.5 | 0.426 | 26.4 | |
| MoCoTraining paradigm=learning general representations2021.11 | 39.7 | 0.39 | 24.2 | |
| DCCMTraining paradigm=without using contrastive learning2021.11 | 32.7 | 0.285 | 17.3 | |
| PICATraining paradigm=without using contrastive learning2021.11 | 32.2 | 0.296 | 15.9 | |
| SupervisedBackbone=ResNet182021.03 | 0.8 | 0.68 | 0.632 | |
| ConNREvaluation Protocol=k-means2023.12 | 0.604 | 60.4 | 44.3 | |
| SPICEBackbone=ResNet342021.03 | 0.584 | 0.583 | 0.422 | |
| ProPosEvaluation Protocol=k-means2023.12 | 0.578 | 58.2 | 42.3 | |
| BYOLEvaluation Protocol=k-means2023.12 | 0.539 | 55.5 | 37.6 | |
| SPICEbackbone=WideResNet-28-8, joint training=true2021.03 | 0.538 | 0.567 | 0.387 | |
| SPICEBackbone=ResNet182021.03 | 0.535 | 0.565 | 0.404 | |
| RUCSCANBackbone=ResNet182021.03 | 0.533 | — | — | |
| TCL2023.12 | 0.531 | 52.9 | 35.7 | |
| DCN + BRBContrastive auxiliary tasks=true, Self-labeling=true2024.11 | 0.5232 | — | — | |
| DCNContrastive auxiliary tasks=true, Self-labeling=true2024.11 | 0.5113 | — | — | |
| SCANBackbone=ResNet182021.03 | 0.507 | 0.486 | 0.333 | |
| SCAN2023.12 | 0.507 | 48.6 | 33.3 | |
| IDEC + BRBContrastive auxiliary tasks=true, Self-labeling=true2024.11 | 0.5 | — | — | |
| TCC2023.12 | 0.491 | 47.9 | 31.2 | |
| IDECContrastive auxiliary tasks=true, Self-labeling=true2024.11 | 0.4885 | — | — | |
| K principal concept identification2021.04 | 0.484 | 0.515 | 0.343 | |
| DECContrastive auxiliary tasks=true, Self-labeling=true2024.11 | 0.4837 | — | — | |
| ConCURLStrategy=max-performance2021.05 | 0.479 | 0.468 | 0.3034 | |
| NNMBackbone=ResNet182021.03 | 0.477 | 0.484 | 0.316 | |
| NMM2023.12 | 0.477 | 48.4 | 31.6 | |
| GCC2023.12 | 0.472 | 47.2 | 30.5 | |
| DEC + BRBContrastive auxiliary tasks=true, Self-labeling=true2024.11 | 0.4718 | — | — | |
| SPICEsBackbone=ResNet182021.03 | 0.468 | 0.457 | 0.321 | |
| SPICEsbackbone=WideResNet-28-8, joint training=false2021.03 | 0.468 | 0.448 | 0.294 | |
| PT+SCAN2021.04 | 0.467 | 0.458 | 0.407 | |
| SCANContrastive auxiliary tasks=true, Self-labeling=true2024.11 | 0.459 | — | — | |
| SCAN_MoCoBackbone=ResNet182021.03 | 0.455 | 0.472 | 0.31 | |
| MICE2023.12 | 0.44 | 43.6 | 28 | |
| DivClust2023.12 | 0.437 | 44 | 28.3 | |
| CCStrategy=max-performance2021.05 | 0.429 | 0.431 | 0.266 | |
| CCbackbone=WideResNet-28-82021.03 | 0.429 | 0.431 | 0.266 | |
| CC2023.12 | 0.429 | 43.1 | 26.6 | |
| IDFDStrategy=max-performance2021.05 | 0.425 | 0.426 | 0.264 | |
| IDFD2021.03 | 0.425 | 0.426 | 0.264 | |
| IDFD2023.12 | 0.425 | 42.6 | 26.4 | |
| PT Only2021.04 | 0.41 | 0.452 | 0.24 | |
| IDStrategy=max-performance2021.05 | 0.409 | 0.392 | 0.243 | |
| TSUCBackbone=ResNet182021.03 | 0.353 | — | — | |
| PICAStrategy=max-performance2021.05 | 0.337 | 0.31 | 0.171 | |
| PICA2021.03 | 0.337 | 0.31 | 0.171 | |
| DCCMStrategy=max-performance2021.05 | 0.327 | 0.285 | 0.173 | |
| DCCM2021.03 | 0.327 | 0.285 | 0.173 | |
| DCCM2023.12 | 0.327 | 28.5 | 17.3 | |
| PICA2023.12 | 0.322 | 29.6 | 15.9 | |
| GATClusterStrategy=max-performance2021.05 | 0.281 | 0.215 | 0.116 | |
| GATCluster2021.03 | 0.281 | 0.215 | 0.116 | |
| IICStrategy=max-performance2021.05 | 0.257 | — | — | |
| IICbackbone=WideResNet-28-82021.03 | 0.257 | — | — | |
| IIC2023.12 | 0.257 | — | — | |
| DSEC2021.03 | 0.255 | 0.212 | 0.11 | |
| DACStrategy=max-performance2021.05 | 0.238 | 0.185 | 0.088 | |
| DAC2021.03 | 0.238 | 0.185 | 0.088 | |
| ZS-Naive2021.04 | 0.208 | 0.25 | 0.043 | |
| Deep ClusterStrategy=max-performance2021.05 | 0.189 | — | — | |
| DeepCluster2021.03 | 0.189 | — | — | |
| DECStrategy=max-performance2021.05 | 0.185 | 0.136 | 0.05 | |
| DEC2021.03 | 0.185 | 0.136 | 0.05 | |
| AEStrategy=max-performance2021.05 | 0.165 | 0.1 | 0.048 | |
| AE2021.03 | 0.165 | 0.1 | 0.048 | |
| ADCStrategy=max-performance2021.05 | 0.16 | — | — | |
| ADCBackbone=ResNet182021.03 | 0.16 | — | — | |
| VAE2021.03 | 0.152 | 0.108 | 0.04 | |
| SDAEStrategy=max-performance2021.05 | 0.151 | 0.111 | 0.046 | |
| SDAE2021.03 | 0.151 | 0.111 | 0.046 | |
| DCGAN2021.03 | 0.151 | 0.12 | 0.045 | |
| ACStrategy=max-performance2021.05 | 0.138 | 0.098 | 0.034 | |
| AC2021.03 | 0.138 | 0.098 | 0.034 | |
| JULEStrategy=max-performance2021.05 | 0.137 | 0.103 | 0.033 | |
| JULE2021.03 | 0.137 | 0.103 | 0.033 | |
| SCStrategy=max-performance2021.05 | 0.136 | 0.09 | 0.022 | |
| SC2021.03 | 0.136 | 0.09 | 0.022 | |
| DeCNNStrategy=max-performance2021.05 | 0.133 | 0.092 | 0.038 | |
| DeCNN2021.03 | 0.133 | 0.092 | 0.038 | |
| k-meansStrategy=max-performance2021.05 | 0.13 | 0.084 | 0.028 | |
| k-means2021.03 | 0.13 | 0.084 | 0.028 | |
| NMFStrategy=max-performance2021.05 | 0.118 | 0.079 | 0.026 | |
| NMF2021.03 | 0.118 | 0.079 | 0.026 | |
| IICTraining paradigm=without using contrastive learning2021.11 | — | 0.257 | — |