Class-Incremental Learning on CUB200
90.5Last AccuracyJoint full tuning
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Joint full tuningBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 90.5 | — | — | |
| MOSBackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 89.91 | — | 92.87 | |
| E2-LoRABackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 89.77 | — | 92.68 | |
| TSVDBackbone=ViT/B-Sup21K2026.01 | 89.23 | — | — | |
| SoTUBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 89.1 | 92.7 | — | |
| RanPACBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 88.9 | 92.6 | — | |
| HiDe-LoRABackbone=ViT/B-Sup21K2026.01 | 88.76 | — | 89.32 | |
| SSIATBackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 88.75 | — | 93 | |
| Joint TrainingBackbone=ViT/B-Sup21K2026.01 | 88.43 | — | — | |
| TUNABackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 88.4 | — | 92.01 | |
| Joint TrainingBackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 88.26 | — | — | |
| RanPACBackbone=ViT/B-Sup21K2026.01 | 88.13 | — | — | |
| RanPACBackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 87.88 | — | 92.57 | |
| NSP-SeqKD + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 87.65 | — | 92.12 | |
| SeqKD + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 87.22 | — | 92.14 | |
| HiDe-PromptBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 87.2 | 87.8 | — | |
| LoRA-SeqKD + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 87.07 | — | 91.63 | |
| LoRA-SeqKD + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 87.07 | — | 91.65 | |
| LoRA-SeqFT + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 86.72 | — | 91.53 | |
| SLCA++Backbone=ViT/B-Sup21K2026.01 | 86.59 | — | 91.63 | |
| ADAM (Ada)Backbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=10, variant=Ada2024.11 | 86.2 | 91.1 | — | |
| CoMABackbone=ViT/B-Sup21K2026.01 | 85.95 | — | 90.75 | |
| RanProj + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 85.73 | — | 90.92 | |
| NSP-SeqFT + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 85.68 | — | 91.45 | |
| ADAM (SSF)Backbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=10, variant=SSF2024.11 | 85.6 | 90.5 | — | |
| NSP-SeqKD + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 85.23 | — | 91.24 | |
| SimpleCILBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 85.2 | 90.6 | — | |
| SLCABackbone=ViT/B-Sup21K2026.01 | 84.71 | — | 90.94 | |
| SLCABackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 84.71 | — | 90.94 | |
| DERBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=10, Examples per class=202024.11 | 84.7 | 88.9 | — | |
| SeqKD + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 84.2 | — | 91.1 | |
| LoRA-SeqFT + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 83.82 | — | 89.74 | |
| MEMOBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=10, Examples per class=202024.11 | 82.7 | 88.9 | — | |
| SeqFT + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 82.27 | — | 89.7 | |
| CoOp + OursBackbone=ViT-B/162024.03 | 81.95 | 86.99 | — | |
| OursBackbone=ViT-B/162024.03 | 81.64 | 86.93 | — | |
| NSP-SeqFT + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 80.62 | — | 89.44 | |
| LAE-AdapterBackbone=ViT/B-Sup21K2026.01 | 80.52 | — | 84.75 | |
| iCaRLBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=10, Examples per class=202024.11 | 80.5 | 87.6 | — | |
| CODA-PBackbone=ViT-B/162024.03 | 80.39 | 84.77 | — | |
| EASEBackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 79.9 | — | 86.65 | |
| PROOFBackbone=ViT-B/162024.03 | 79.35 | 83.98 | — | |
| CODA-PromptBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 79.3 | 84.1 | — | |
| Single-task JOINTBackbone=ViT-B/162024.03 | 79.23 | 82.25 | — | |
| AttriCLIP + OursBackbone=ViT-B/162024.03 | 79.01 | 83.71 | — | |
| ESNBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 78.4 | 82.3 | — | |
| FOSTERBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=10, Examples per class=202024.11 | 76.7 | 81.8 | — | |
| DualPromptBackbone=ViT-B/162024.03 | 76.29 | 82.37 | — | |
| iCaRLBackbone=ViT-B/162024.03 | 75.1 | 82.39 | — | |
| MaPLe + OursBackbone=ViT-B/162024.03 | 74.24 | 81.53 | — | |
| CODA-PromptBackbone=ViT/B-Sup21K2026.01 | 71.43 | — | 78.61 | |
| SeqFT + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 70.94 | — | 82.55 | |
| DualPromptBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 68.7 | 79.8 | — | |
| CoOpBackbone=ViT-B/162024.03 | 68.53 | 76.62 | — | |
| L2PBackbone=ViT-B/162024.03 | 68.47 | 76.98 | — | |
| L2PBackbone=ViT-B/16, Pre-training=ImageNet21K, Tasks=102024.11 | 68.1 | 79.6 | — | |
| DualPromptBackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 66 | — | 77.92 | |
| MaPLeBackbone=ViT-B/162024.03 | 64.43 | 73.38 | — | |
| L2PBackbone=ViT-B/16-IN21K, Number of tasks=102026.05 | 62.21 | — | 73.83 | |
| Ours w/o VIBackbone=ViT-B/162024.03 | 61.87 | 72.21 | — | |
| VPTBackbone=ViT-B/162024.03 | 60.03 | 69.14 | — | |
| Continual-CLIPBackbone=ViT-B/162024.03 | 54.8 | 67 | — | |
| CLIP-AdapterBackbone=ViT-B/162024.03 | 54.49 | 67.41 | — | |
| AttriCLIPBackbone=ViT-B/162024.03 | 52.12 | 65.26 | — |