Image Classification on CIFAR100 (Transfer/Average/Last Scores)
68.2AccuracyOurs (T*csg)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Ours (T*csg)Backbone=Eff. Net b52025.04 | 68.2 | — | — | — | — | — | — | — | |
| Ours (T*cn)Backbone=Eff. Net b52025.04 | 67.81 | — | — | — | — | — | — | — | |
| Ours (T*)Backbone=Eff. Net b52025.04 | 66.05 | — | — | — | — | — | — | — | |
| Ours (T*csgcn)Backbone=Eff. Net b52025.04 | 65.93 | — | — | — | — | — | — | — | |
| T = √MBackbone=RegNet 8 GF2025.04 | 65.16 | — | — | — | — | — | — | — | |
| Ours (T*)Backbone=RegNet 8 GF2025.04 | 64.94 | — | — | — | — | — | — | — | |
| Ours (T*csg)Backbone=RegNet 8 GF2025.04 | 63.6 | — | — | — | — | — | — | — | |
| Ours (T*csgcn)Backbone=RegNet 8 GF2025.04 | 63.56 | — | — | — | — | — | — | — | |
| Ours (T*cn)Backbone=RegNet 8 GF2025.04 | 63.48 | — | — | — | — | — | — | — | |
| T = 1Backbone=Eff. Net b52025.04 | 62.34 | — | — | — | — | — | — | — | |
| T = √MBackbone=ResNet502025.04 | 60.99 | — | — | — | — | — | — | — | |
| Ours (T*csgcn)Backbone=Eff. Net b02025.04 | 60.14 | — | — | — | — | — | — | — | |
| Ours (T*)Backbone=ResNet502025.04 | 59.95 | — | — | — | — | — | — | — | |
| Ours (T*csg)Backbone=Eff. Net b02025.04 | 59.2 | — | — | — | — | — | — | — | |
| Ours (T*)Backbone=Eff. Net b02025.04 | 59.01 | — | — | — | — | — | — | — | |
| Ours (T*cn)Backbone=Eff. Net b02025.04 | 58.94 | — | — | — | — | — | — | — | |
| T = 1Backbone=Eff. Net b02025.04 | 58.75 | — | — | — | — | — | — | — | |
| Ours (T*csg)Backbone=ResNet502025.04 | 58.7 | — | — | — | — | — | — | — | |
| Insert LNBackbone=Eff. Net b02025.04 | 58.48 | — | — | — | — | — | — | — | |
| Ours (T*cn)Backbone=ResNet502025.04 | 57.93 | — | — | — | — | — | — | — | |
| Ours (T*csgcn)Backbone=ResNet502025.04 | 57.89 | — | — | — | — | — | — | — | |
| T = 1Backbone=RegNet 8 GF2025.04 | 56.52 | — | — | — | — | — | — | — | |
| Insert LNBackbone=Eff. Net b52025.04 | 55.5 | — | — | — | — | — | — | — | |
| Insert LNBackbone=RegNet 8 GF2025.04 | 55.41 | — | — | — | — | — | — | — | |
| T = 1Backbone=ConvNeXt-S2025.04 | 50.34 | — | — | — | — | — | — | — | |
| Insert LNBackbone=ResNet502025.04 | 49.64 | — | — | — | — | — | — | — | |
| Insert LNBackbone=ConvNeXt-S2025.04 | 49.58 | — | — | — | — | — | — | — | |
| T = 1Backbone=ResNet502025.04 | 49.18 | — | — | — | — | — | — | — | |
| T = √MBackbone=Eff. Net b02025.04 | 46.23 | — | — | — | — | — | — | — | |
| Ours (T*csgcn)Backbone=ConvNeXt-S2025.04 | 42.7 | — | — | — | — | — | — | — | |
| Ours (T*csg)Backbone=ConvNeXt-S2025.04 | 41.7 | — | — | — | — | — | — | — | |
| Ours (T*cn)Backbone=ConvNeXt-S2025.04 | 41 | — | — | — | — | — | — | — | |
| Ours (T*)Backbone=ConvNeXt-S2025.04 | 35.25 | — | — | — | — | — | — | — | |
| T = √MBackbone=ConvNeXt-S2025.04 | 32.62 | — | — | — | — | — | — | — | |
| T = √MBackbone=Eff. Net b52025.04 | 15.97 | — | — | — | — | — | — | — | |
| AttriCLIPPrompt=text2026.04 | — | — | — | — | 81.4 | 82.3 | 0.9 | — | |
| CLIBBuffer=5002026.04 | — | — | — | 67.16 | — | — | — | 69.68 | |
| CLIBBuffer=20002026.04 | — | — | — | 72.09 | — | — | — | 71.53 | |
| CODA-P-1Prompt=visual2026.04 | — | — | — | — | 72.3 | 66.9 | -5.4 | — | |
| ConDU (FT)Adaptation strategy=Fine-tuning2025.03 | — | 69.5 | 74.1 | 75 | — | — | — | — | |
| ConDU (LoRA)Adaptation strategy=LoRA2025.03 | — | 68.5 | 75.4 | 76.7 | — | — | — | — | |
| Continual FTAdaptation strategy=Continual Fine-tuning2025.03 | — | 53 | 59.1 | 60.1 | — | — | — | — | |
| Continual-CLIPPrompt=text2026.04 | — | — | — | — | 73.4 | 73.4 | 0 | — | |
| CoOp-1 (1000)Prompt=text2026.04 | — | — | — | — | 67.6 | 61.1 | -6.5 | — | |
| DER++Buffer=5002026.04 | — | — | — | 65.63 | — | — | — | 66.92 | |
| DER++Buffer=20002026.04 | — | — | — | 65.68 | — | — | — | 69.42 | |
| DualPromptBuffer=02026.04 | — | — | — | 56.82 | — | — | — | 67.07 | |
| DualPrompt-1Prompt=visual2026.04 | — | — | — | — | 54 | 49.5 | -4.5 | — | |
| ERBuffer=5002026.04 | — | — | — | 60.68 | — | — | — | 65.57 | |
| ERBuffer=20002026.04 | — | — | — | 71.81 | — | — | — | 69.86 | |
| ER-ACEBuffer=5002026.04 | — | — | — | 72.07 | — | — | — | 69.36 | |
| ER-ACEBuffer=20002026.04 | — | — | — | 74.75 | — | — | — | 70.59 | |
| EWCBuffer=02026.04 | — | — | — | 52.83 | — | — | — | 49.51 | |
| FinetuningBuffer=02026.04 | — | — | — | 10.42 | — | — | — | 19.71 | |
| HiDe-Prompt-1Prompt=visual2026.04 | — | — | — | — | 81 | 77.2 | -3.8 | — | |
| Individual FTAdaptation strategy=Individual Fine-tuning2025.03 | — | — | 76.8 | — | — | — | — | — | |
| Insert LNBackbone=Avg.2025.04 | — | — | 53.72 | — | — | — | — | — | |
| L2PBuffer=02026.04 | — | — | — | 41.63 | — | — | — | 57.08 | |
| L2P-1Prompt=visual2026.04 | — | — | — | — | 49.4 | 45.7 | -3.7 | — | |
| Linear ProbeBuffer=02026.04 | — | — | — | 23.07 | — | — | — | 49.69 | |
| LwFBuffer=02026.04 | — | — | — | 36.53 | — | — | — | 55.51 | |
| MISABuffer=02026.04 | — | — | — | 80.98 | — | — | — | 80.55 | |
| MISABuffer=5002026.04 | — | — | — | 82.27 | — | — | — | 82.37 | |
| MISABuffer=20002026.04 | — | — | — | 85.32 | — | — | — | 83.58 | |
| MoEContinual learning method=MoE2025.03 | — | 68.2 | 73.9 | 74.9 | — | — | — | — | |
| MVPBuffer=02026.04 | — | — | — | 62.59 | — | — | — | 68.1 | |
| MVPBuffer=5002026.04 | — | — | — | 79.32 | — | — | — | 76.06 | |
| MVPBuffer=20002026.04 | — | — | — | 84.42 | — | — | — | 78.65 | |
| Ours (T*)Backbone=Avg.2025.04 | — | — | 57.04 | — | — | — | — | — | |
| Ours (T*cn)Backbone=Avg.2025.04 | — | — | 57.83 | — | — | — | — | — | |
| Ours (T*csg)Backbone=Avg.2025.04 | — | — | 58.28 | — | — | — | — | — | |
| Ours (T*csgcn)Backbone=Avg.2025.04 | — | — | 58.04 | — | — | — | — | — | |
| ProTPSPrompt=text∗2026.04 | — | — | — | — | 83.3 | 84.4 | 1.1 | — | |
| RMBuffer=5002026.04 | — | — | — | 23.94 | — | — | — | 40.86 | |
| RMBuffer=20002026.04 | — | — | — | 65.51 | — | — | — | 53.27 | |
| S-liPromptsPrompt=v + t2026.04 | — | — | — | — | 58.9 | 53.3 | -5.6 | — | |
| SinglePromptBuffer=02026.04 | — | — | — | 87.53 | — | — | — | 85.58 | |
| SinglePromptBuffer=5002026.04 | — | — | — | 85.61 | — | — | — | 84.34 | |
| SinglePromptBuffer=20002026.04 | — | — | — | 88.34 | — | — | — | 86.07 | |
| T = √MBackbone=Avg.2025.04 | — | — | 44.19 | — | — | — | — | — | |
| T = 1Backbone=Avg.2025.04 | — | — | 55.43 | — | — | — | — | — | |
| WiSE-FTAdaptation strategy=WiSE-FT2025.03 | — | 60 | 61 | 59.6 | — | — | — | — | |
| Zero-shotAdaptation strategy=Zero-shot2025.03 | — | — | 68.2 | — | — | — | — | — | |
| Zero-Shot CLIPPrompt=text2026.04 | — | — | — | — | 78.3 | 78.3 | 0 | — | |
| ZSCLContinual learning method=ZSCL2025.03 | — | 68.1 | 66.5 | 63.7 | — | — | — | — |