Image Clustering on ImageNet-10
0.996NMILAIC framework
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LAIC frameworkBackbone=ViT-B/322026.03 | 0.996 | 99.8 | 0.997 | |
| GSEC2026.05 | 0.996 | 99.8 | 0.996 | |
| GradNorm2025.10 | 0.987 | 99.4 | 0.987 | |
| GradNormBackbone=ViT-B/322026.03 | 0.987 | 99.4 | 0.987 | |
| GradNorm2026.05 | 0.987 | 99.4 | 0.987 | |
| TACtrained=true2023.10 | 0.985 | 99.2 | 0.983 | |
| TACBackbone=ViT-B/322026.03 | 0.985 | 99.2 | 0.983 | |
| TAC2026.05 | 0.985 | 99.2 | 0.983 | |
| CAE2026.05 | 0.981 | 98.8 | 0.974 | |
| PRO-DSCBackbone=ViT-B/322026.03 | 0.98 | 99 | 0.978 | |
| TACtrained=false2023.10 | 0.975 | 98.6 | 0.97 | |
| TAC2025.10 | 0.975 | 98.6 | 0.97 | |
| SIC2023.10 | 0.97 | 98.2 | 0.961 | |
| SIC2025.10 | 0.97 | 98.2 | 0.961 | |
| SICBackbone=ViT-B/322026.03 | 0.97 | 98.2 | 0.961 | |
| SIC2026.05 | 0.97 | 98.2 | 0.961 | |
| CLIPEvaluation Protocol=k-means2023.10 | 0.969 | 98.2 | 0.961 | |
| CLIPmode=k-means2025.10 | 0.969 | 98.2 | 0.961 | |
| CLIPalgorithm=k-means2026.05 | 0.969 | 98.2 | 0.961 | |
| CLIPEvaluation Protocol=zero-shot2023.10 | 0.958 | 97.6 | 0.949 | |
| CLIPmode=zero-shot2025.10 | 0.958 | 97.6 | 0.949 | |
| ZS-CLIPBackbone=ViT-B/32, zero-shot=true2026.03 | 0.958 | 97.6 | 0.949 | |
| SPICEextra training data=true2022.10 | 0.927 | 96.9 | 0.933 | |
| DPACOPASelf-labeling fine-tuning=true, Clustering mode=Online2024.07 | 0.925 | 97 | 0.935 | |
| DMICCBackbone=ResNet2026.03 | 0.917 | 96.2 | 0.916 | |
| ConNREvaluation Protocol=k-means2023.12 | 0.911 | 96.4 | 0.922 | |
| DPACOPAClustering mode=Online2024.07 | 0.908 | 96.2 | 0.918 | |
| ConCURLStrategy=max-performance2021.05 | 0.907 | 95.8 | 0.909 | |
| SPICEbackbone=ResNet-34, joint training=true2021.03 | 0.902 | 95.9 | 0.912 | |
| SPICESelf-labeling fine-tuning=true, Clustering mode=Online2024.07 | 0.902 | 95.9 | 0.912 | |
| DPACpacClustering mode=Offline2024.07 | 0.901 | 96 | 0.911 | |
| IDFD2021.05 | 0.898 | 95.4 | 0.901 | |
| IDFDStrategy=max-performance2021.05 | 0.898 | 95.4 | 0.901 | |
| IDFDTraining paradigm=improving representation learning2021.11 | 0.898 | 95.4 | 0.901 | |
| IDFD2021.03 | 0.898 | 95.4 | 0.901 | |
| IDFD2023.12 | 0.898 | 95.1 | 0.901 | |
| IDFDClustering mode=Offline2024.07 | 0.898 | 95.4 | 0.901 | |
| IDFD2023.10 | 0.898 | 95.4 | 0.901 | |
| IDFD2025.10 | 0.898 | 95.4 | 0.901 | |
| ProPosTraining paradigm=improving representation learning2021.11 | 0.896 | 95.6 | 0.906 | |
| ProPosEvaluation Protocol=k-means2023.12 | 0.896 | 95.6 | 0.906 | |
| DivClust CDiversity Target (DT)=0.952023.04 | 0.891 | 93.6 | 0.878 | |
| DivClust2023.12 | 0.891 | 93.6 | 0.878 | |
| BYOLBackbone=ResNet2026.03 | 0.884 | 94.7 | 0.889 | |
| DeepCluE2023.04 | 0.882 | 92.4 | 0.856 | |
| KBCPoints=13,000, Dim=128, Clusters=102026.02 | 0.88 | — | — | |
| DivClust CDiversity Target (DT)=0.92023.04 | 0.879 | 92 | 0.859 | |
| DivClust CDiversity Target (DT)=0.82023.04 | 0.879 | 91.8 | 0.851 | |
| TCLTraining paradigm=directly outputting cluster assignments, Image size=224x2242021.11 | 0.875 | 89.5 | 0.837 | |
| TCL2023.12 | 0.875 | 89.5 | 0.837 | |
| TCLSelf-labeling fine-tuning=true, Clustering mode=Online2024.07 | 0.875 | 89.5 | 0.837 | |
| TCL2025.10 | 0.875 | 89.5 | 0.837 | |
| TCLBackbone=ResNet2026.03 | 0.875 | 89.5 | 0.837 | |
| TCL2026.05 | 0.875 | 89.5 | 0.837 | |
| IDFO2021.05 | 0.871 | 94.2 | 0.876 | |
| TCLICHclustering=k-means on ICH features2022.10 | 0.869 | 89.1 | 0.823 | |
| TCL2022.10 | 0.868 | 92.9 | 0.837 | |
| IDTemperature Setting=tuned2021.05 | 0.867 | 93.7 | 0.865 | |
| IDStrategy=max-performance2021.05 | 0.867 | 93.7 | 0.865 | |
| BYOLTraining paradigm=learning general representations2021.11 | 0.866 | 93.9 | 0.872 | |
| BYOLEvaluation Protocol=k-means2023.12 | 0.866 | 93.9 | 0.872 | |
| DivClust CDiversity Target (DT)=1.2023.04 | 0.86 | 89.5 | 0.825 | |
| CCStrategy=max-performance2021.05 | 0.859 | 89.3 | 0.822 | |
| CC2022.10 | 0.859 | 92.7 | 0.87 | |
| CCTraining paradigm=directly outputting cluster assignments, Image size=224x2242021.11 | 0.859 | 89.3 | 0.822 | |
| CCbackbone=ResNet-342021.03 | 0.859 | 89.3 | 0.822 | |
| CC2023.04 | 0.859 | 89.3 | 0.822 | |
| CC2023.12 | 0.859 | 89.3 | 0.822 | |
| CCClustering mode=Online, Cluster size constraint=true2024.07 | 0.859 | 89.3 | 0.822 | |
| CC2023.10 | 0.859 | 89.3 | 0.822 | |
| CC2025.10 | 0.859 | 89.3 | 0.822 | |
| CCBackbone=ResNet2026.03 | 0.859 | 89.3 | 0.822 | |
| LFSSBackbone=ResNet2026.03 | 0.856 | 93.2 | 0.857 | |
| LFSS2026.05 | 0.856 | 93.2 | 0.857 | |
| DivClust CDiversity Target (DT)=0.72023.04 | 0.85 | 90 | 0.819 | |
| DivClust2025.10 | 0.85 | 90 | 0.819 | |
| DivClust2026.05 | 0.85 | 90 | 0.819 | |
| TCCTraining paradigm=directly outputting cluster assignments2021.11 | 0.848 | 89.7 | 0.825 | |
| TCC2023.12 | 0.848 | 89.7 | 0.825 | |
| TCCClustering mode=Online, Cluster size constraint=true2024.07 | 0.848 | 89.7 | 0.825 | |
| TCC2023.10 | 0.848 | 89.7 | 0.825 | |
| TCC2025.10 | 0.848 | 89.7 | 0.825 | |
| TCCBackbone=ResNet2026.03 | 0.848 | 89.7 | 0.825 | |
| TCC2026.05 | 0.848 | 89.7 | 0.825 | |
| GCCTraining paradigm=directly outputting cluster assignments2021.11 | 0.842 | 90.1 | 0.822 | |
| GCC2023.12 | 0.842 | 90.1 | 0.822 | |
| GCCClustering mode=Online, Cluster size constraint=true2024.07 | 0.842 | 90.1 | 0.822 | |
| GCC2023.10 | 0.842 | 90.1 | 0.822 | |
| GCC2025.10 | 0.842 | 90.1 | 0.822 | |
| GCCBackbone=ResNet2026.03 | 0.842 | 90.1 | 0.822 | |
| GCC2026.05 | 0.842 | 90.1 | 0.822 | |
| PCL2022.10 | 0.841 | 90.7 | 0.822 | |
| PCLTraining paradigm=improving representation learning2021.11 | 0.841 | 90.7 | 0.822 | |
| PCLClustering mode=Offline2024.07 | 0.841 | 90.7 | 0.822 | |
| SimSiamTraining paradigm=learning general representations2021.11 | 0.831 | 92.1 | 0.833 | |
| CRLC2025.10 | 0.831 | 85.4 | 0.759 | |
| CRLC2026.05 | 0.831 | 85.4 | 0.759 | |
| RPSC2025.10 | 0.83 | 92.7 | 0.858 | |
| RPSC2026.05 | 0.83 | 92.7 | 0.858 | |
| SPICEsbackbone=ResNet-34, joint training=false2021.03 | 0.828 | 92.1 | 0.836 |