Image Classification on ImageNet-100
91.3AccuracyVisualAtom-21k
Evaluation results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| VisualAtom-21kModel=ViT-B, Type=FDSL2023.03 | 91.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisualAtom-1kModel=ViT-T, Type=FDSL2023.03 | 90.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RCDB-21kModel=ViT-B, Type=FDSL2023.03 | 90.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PromptSRC + Bi-CoGBackbone=ViT-B/162025.10 | 90.15 | — | — | — | — | — | — | — | — | — | — | — | — | 91.07 | — | — | |
| CoOp + Bi-CoGBackbone=ViT-B/162025.10 | 89.59 | — | — | — | — | — | — | — | — | — | — | — | — | 90.39 | — | — | |
| MaPLe + Bi-CoGBackbone=ViT-B/162025.10 | 89.52 | — | — | — | — | — | — | — | — | — | — | — | — | 90.41 | — | — | |
| PromptSRCBackbone=ViT-B/162025.10 | 89.43 | — | — | — | — | — | — | — | — | — | — | — | — | 90.49 | — | — | |
| MaPLeBackbone=ViT-B/162025.10 | 89.07 | — | — | — | — | — | — | — | — | — | — | — | — | 90 | — | — | |
| ImageNet-1kModel=ViT-T, Type=SSL (DINO)2023.03 | 89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RCDB-1kModel=ViT-T, Type=FDSL2023.03 | 88.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FractalDB-1kModel=ViT-T, Type=FDSL2023.03 | 88.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ExFractalDB-1kModel=ViT-T, Type=FDSL2023.03 | 88.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LRW-Hard2024.03 | 87.95 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DTTN†-SResolution=224^22025.02 | 87.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LRWOpt2024.03 | 87.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAPLE2024.03 | 87.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LRW-Random2024.03 | 87.62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEDownstream Task=Fine-tuning, Backbone=ViT-Large2022.10 | 87.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| U-MAEDownstream Task=Fine-tuning, Backbone=ViT-Large2022.10 | 87.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BILAW2024.03 | 87.25 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MONet-TResolution=224^22025.02 | 87.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FSR2024.03 | 87.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SparseSwin with L2#Params(100 class)=17.58 M, Input Resolution=224^2, Model Type=Transformer2023.09 | 86.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RHO-LOSS2024.03 | 86.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ERM2024.03 | 86.91 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEDownstream Task=Fine-tuning, Backbone=ViT-Base2022.10 | 86.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SparseSwinArchitecture=SparseSwin, Pre-training Dataset=ImageNet-1K (1.3M), Evaluation Protocol=Fine-tuned2025.02 | 86.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LRW-Easy2024.03 | 86.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| U-MAEDownstream Task=Fine-tuning, Backbone=ViT-Base2022.10 | 86.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Fixed PrototypesLoss=SCL, Passes=N, Backbone=ResNet502026.05 | 86.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MWN2024.03 | 86.78 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CLIPBackbone=ViT-B/162025.10 | 86.5 | — | — | — | — | — | — | — | — | — | — | — | — | 88.33 | — | — | |
| MBW2024.03 | 86.34 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| D-PolyNetsResolution=224^22025.02 | 86.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hybrid Π-NetsResolution=224^22025.02 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Resnet18Resolution=224^22025.02 | 85.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Swin-T#Params(100 class)=27.6 M, Input Resolution=224^2, Model Type=Transformer2023.09 | 85.22 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CTSNArchitecture=SEW-ResNet34, T=42026.01 | 85.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NONLLoss=NONL, Backbone=ResNet502026.05 | 84.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SCOTT-12/16Architecture=SCOTT-12/16, Pre-training Dataset=Target dataset (unlabeled), Evaluation Protocol=Frozen2025.02 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Linear ProbingLoss=SCL, Passes=T × N, Backbone=ResNet502026.05 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Normalized Linear ProbingLoss=SCL, Passes=T × N, Backbone=ResNet502026.05 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Res-MLPResolution=224^22025.02 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NTCELoss=NTCE, Backbone=ResNet502026.05 | 84.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ETF + DRLoss=ETF + DR, Backbone=ResNet502026.05 | 84.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CELoss=CE, Backbone=ResNet502026.05 | 84.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NORMFACELoss=NORMFACE, Backbone=ResNet502026.05 | 84.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MLP-MixerResolution=224^22025.02 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CTSNArchitecture=ResNet34, T=42026.01 | 83.78 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FPTArchitecture=SEW-ResNet34, T=42026.01 | 83.27 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PASSModel=ViT-T, Type=SSL (DINO)2023.03 | 82.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WONN (Ch = 256 → 256)S(θi)&I(θi) configuration type=as MLPs, Channel configuration (Ch)=256 → 256, # Parameters=12.05M2026.05 | 82.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PDCResolution=224^22025.02 | 82.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WONN (Ch = 64 → 256)S(θi)&I(θi) configuration type=as trigonometric functions, Channel configuration (Ch)=64 → 256, # Parameters=7.43M2026.05 | 82.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WONN (Ch = 256 → 256)S(θi)&I(θi) configuration type=as trigonometric functions, Channel configuration (Ch)=256 → 256, # Parameters=11.87M2026.05 | 82.22 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WONN (Ch = 64 → 256)S(θi)&I(θi) configuration type=as MLPs, Channel configuration (Ch)=64 → 256, # Parameters=7.56M2026.05 | 82.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WONN (Ch = 128 → 128)S(θi)&I(θi) configuration type=as trigonometric functions, Channel configuration (Ch)=128 → 128, # Parameters=3.00M2026.05 | 81.76 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LocalZOArchitecture=SEW-ResNet34, T=42026.01 | 81.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| WONN (Ch = 128 → 128)S(θi)&I(θi) configuration type=as MLPs, Channel configuration (Ch)=128 → 128, # Parameters=3.09M2026.05 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Π-NetsResolution=224^22025.02 | 81.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-50# Parameters=23.71M2026.05 | 81.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoOpBackbone=ViT-B/162025.10 | 81.17 | — | — | — | — | — | — | — | — | — | — | — | — | 81.97 | — | — | |
| SCOTT-7/16Architecture=SCOTT-7/16, Pre-training Dataset=Target dataset (unlabeled), Evaluation Protocol=Frozen2025.02 | 81.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViT-B#Params(100 class)=85.87 M, Input Resolution=224^2, Model Type=Transformer2023.09 | 80.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IMP+LTSArchitecture=SEW-ResNet34, T=42026.01 | 80.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=10, Evaluation Protocol=Linear Probing2026.04 | 80.76 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=0, Evaluation Protocol=Linear Probing2026.04 | 80.62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=20, Evaluation Protocol=Linear Probing2026.04 | 80.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=30, Evaluation Protocol=Linear Probing2026.04 | 80.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=40, Evaluation Protocol=Linear Probing2026.04 | 80.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=50, Evaluation Protocol=Linear Probing2026.04 | 80.42 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=60, Evaluation Protocol=Linear Probing2026.04 | 80.42 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=80, Evaluation Protocol=Linear Probing2026.04 | 80.28 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=70, Evaluation Protocol=Linear Probing2026.04 | 80.14 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AKOrNattn# Parameters=4.62M2026.05 | 80.08 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CCARNoise (%)=90, Evaluation Protocol=Linear Probing2026.04 | 79.98 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=0, Evaluation Protocol=Linear Probing2026.04 | 79.84 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=50, Evaluation Protocol=Linear Probing2026.04 | 79.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=60, Evaluation Protocol=Linear Probing2026.04 | 79.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=20, Evaluation Protocol=Linear Probing2026.04 | 79.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=10, Evaluation Protocol=Linear Probing2026.04 | 79.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=90, Evaluation Protocol=Linear Probing2026.04 | 79.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=70, Evaluation Protocol=Linear Probing2026.04 | 79.44 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=30, Evaluation Protocol=Linear Probing2026.04 | 79.36 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=40, Evaluation Protocol=Linear Probing2026.04 | 79.34 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DLMEEvaluation Protocol=linear-test2022.07 | 79.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DLME(ResNet-50, linear)Input Resolution=224^2, Model Type=Convolution2023.09 | 79.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Contrastive (CL)Noise (%)=80, Evaluation Protocol=Linear Probing2026.04 | 79.28 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=0, Evaluation Protocol=Linear Probing2026.04 | 79.14 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DLME-A2Evaluation Protocol=linear-test2022.07 | 79.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=30, Evaluation Protocol=Linear Probing2026.04 | 79.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=10, Evaluation Protocol=Linear Probing2026.04 | 78.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=70, Evaluation Protocol=Linear Probing2026.04 | 78.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=20, Evaluation Protocol=Linear Probing2026.04 | 78.68 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=60, Evaluation Protocol=Linear Probing2026.04 | 78.64 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Diffusion-Driven SelectionIPC=100, Teacher Model=ResNet-18, Synthesis Model Backbone=Stable Diffusion2024.12 | 78.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=50, Evaluation Protocol=Linear Probing2026.04 | 78.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=80, Evaluation Protocol=Linear Probing2026.04 | 78.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NCLNoise (%)=40, Evaluation Protocol=Linear Probing2026.04 | 78.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BYOLEvaluation Protocol=linear-test2022.07 | 78.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |