Class-incremental learning on CIFAR-100 Split (test)
94.76Avg AccNoRGa
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| NoRGaPTM=iBOT-21K2024.05 | 94.76 | 1.34 | 95.86 | |
| NoRGaPTM=Sup-21K2024.05 | 94.48 | 1.44 | 95.83 | |
| NoRGaPTM=iBOT-1K2024.05 | 94.01 | 1.61 | 95.11 | |
| NoRGaPTM=MoCo-1K2024.05 | 93.52 | 1.63 | 94.94 | |
| HiDe-PromptPTM=iBOT-1K2024.05 | 93.48 | 1.63 | 95.02 | |
| NoRGaPTM=DINO-1K2024.05 | 93.43 | 1.65 | 94.65 | |
| HiDe-PromptPTM=iBOT-21K2024.05 | 93.02 | 1.26 | 94.56 | |
| HiDe-PromptPTM=Sup-21K2024.05 | 92.61 | 1.5 | 94.03 | |
| HiDe-PromptPTM=DINO-1K2024.05 | 92.51 | 1.67 | 94.25 | |
| HiDe-PromptPTM=MoCo-1K2024.05 | 91.57 | 1.51 | 93.7 | |
| Upper bound2021.12 | 90.85 | — | — | |
| Open-InCABackbone=ViT-L/162023.03 | 88.3 | 7.1 | — | |
| S-PromptPTM=Sup-21K2024.05 | 87.57 | 3.63 | 91.38 | |
| DualPromptPTM=Sup-21K2024.05 | 87.3 | 3.87 | 91.23 | |
| CODA-PromptPTM=Sup-21K2024.05 | 86.94 | 4.04 | 91.57 | |
| L2P-RBuffer size=50/class2021.12 | 86.31 | 5.83 | — | |
| Open-InCABackbone=ViT-H/142023.03 | 86.1 | 8.2 | — | |
| L2P-RBuffer size=10/class2021.12 | 84.21 | 7.72 | — | |
| DER++Buffer size=50/class2021.12 | 83.94 | 14.55 | — | |
| L2PBuffer size=02021.12 | 83.83 | 7.63 | — | |
| L2P2023.03 | 83.8 | 7.6 | — | |
| L2PPTM=Sup-21K2024.05 | 83.06 | 5.61 | 88.27 | |
| Open-InCABackbone=ViT-B/162023.03 | 83 | 9.1 | — | |
| ERBuffer size=50/class2021.12 | 82.53 | 16.46 | — | |
| Co2LBuffer size=50/class2021.12 | 82.49 | 17.48 | — | |
| GDumbBuffer size=50/class2021.12 | 81.67 | — | — | |
| BiCBuffer size=50/class2021.12 | 81.42 | 17.31 | — | |
| CODA-PromptPTM=iBOT-21K2024.05 | 80.83 | 7.5 | 87.02 | |
| S-PromptPTM=iBOT-21K2024.05 | 79.14 | 8.23 | 85.85 | |
| L2PPTM=iBOT-21K2024.05 | 79.13 | 7.5 | 85.13 | |
| CODA-PromptPTM=iBOT-1K2024.05 | 79.11 | 7.69 | 86.21 | |
| DualPromptPTM=iBOT-21K2024.05 | 78.84 | 8.84 | 86.16 | |
| CODA-PromptPTM=MoCo-1K2024.05 | 77.83 | 12.6 | 84.97 | |
| DualPromptPTM=MoCo-1K2024.05 | 77.56 | 6.54 | 84.37 | |
| CODA-PromptPTM=DINO-1K2024.05 | 77.5 | 8.1 | 84.81 | |
| L2PPTM=MoCo-1K2024.05 | 77.24 | 5.57 | 83.73 | |
| S-PromptPTM=MoCo-1K2024.05 | 77.2 | 7 | 84.47 | |
| S-PromptPTM=iBOT-1K2024.05 | 76.6 | 8.6 | 82.89 | |
| DualPromptPTM=iBOT-1K2024.05 | 76.21 | 9.89 | 83.54 | |
| L2PPTM=iBOT-1K2024.05 | 75.51 | 6.8 | 82.53 | |
| S-PromptPTM=DINO-1K2024.05 | 74.39 | 9.07 | 81.6 | |
| DualPromptPTM=DINO-1K2024.05 | 73.95 | 9.32 | 81.85 | |
| L2PPTM=DINO-1K2024.05 | 72.23 | 8.37 | 79.71 | |
| Co2LBuffer size=10/class2021.12 | 72.15 | 28.55 | — | |
| MULTITASKProtocol=Standard Continual Learning, Replay-buffer (B)=false2023.06 | 70.32 | — | — | |
| ERBuffer size=10/class2021.12 | 67.87 | 33.33 | — | |
| GDumbBuffer size=10/class2021.12 | 67.14 | — | — | |
| BiCBuffer size=10/class2021.12 | 66.11 | 35.24 | — | |
| DER++Buffer size=10/class2021.12 | 61.06 | 39.87 | — | |
| LwF2023.03 | 60.7 | 27.8 | — | |
| LwFBuffer size=02021.12 | 60.69 | 27.77 | — | |
| EWCBuffer size=02021.12 | 47.01 | 33.27 | — | |
| EWC2023.03 | 47 | 33.3 | — | |
| FT-seqBuffer size=02021.12 | 33.61 | 86.87 | — | |
| Full-FT-sequentialtraining_mode=sequential full fine-tuning2023.03 | 33.6 | 86.9 | — | |
| KD w/ BUFFERProtocol=Standard Continual Learning, Replay-buffer (B)=true2023.06 | 25.19 | 40.53 | — | |
| KD w/ QDIProtocol=Standard Continual Learning, Replay-buffer (B)=false2023.06 | 22.53 | 16.5 | — | |
| DER++Protocol=Standard Continual Learning, Replay-buffer (B)=true2023.06 | 21.38 | 56.17 | — | |
| KD w/ BUFFERProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=true2023.06 | 21 | 38.49 | — | |
| DERProtocol=Standard Continual Learning, Replay-buffer (B)=true2023.06 | 19.01 | 65.06 | — | |
| FT-seq-frozenBuffer size=02021.12 | 17.72 | 59.09 | — | |
| LP-sequentialtraining_mode=sequential linear probing2023.03 | 17.7 | 59.1 | — | |
| KDProtocol=Standard Continual Learning, Replay-buffer (B)=false2023.06 | 16.77 | 80.65 | — | |
| KD w/ QDIProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=false2023.06 | 15.86 | 32.73 | — | |
| KDProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=false2023.06 | 12.94 | 58.89 | — | |
| DER++Protocol=Heterogeneous Continual Learning, Replay-buffer (B)=true2023.06 | 10.97 | 73.62 | — | |
| DERProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=true2023.06 | 10.11 | 80.92 | — | |
| ERProtocol=Standard Continual Learning, Replay-buffer (B)=true2023.06 | 9.65 | 85.2 | — | |
| ERProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=true2023.06 | 7.43 | 82.85 | — | |
| DIProtocol=Standard Continual Learning, Replay-buffer (B)=false2023.06 | 7.21 | 63.9 | — | |
| FINETUNEProtocol=Standard Continual Learning, Replay-buffer (B)=false2023.06 | 6.93 | 60.53 | — | |
| DIProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=false2023.06 | 6.2 | 73.7 | — | |
| FINETUNEProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=false2023.06 | 5.27 | 72.99 | — | |
| A-GEMProtocol=Standard Continual Learning, Replay-buffer (B)=true2023.06 | 5.04 | 91.93 | — | |
| A-GEMProtocol=Heterogeneous Continual Learning, Replay-buffer (B)=true2023.06 | 4.62 | 88.78 | — |