Class-Incremental Learning on CUB-200, Cars-196, CIFAR-100, ImageNet-R
88.05Last AccuracyJoint Training
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Joint TrainingBackbone=ViT/B-Sup21K2026.01 | 88.05 | — | |
| SeqKD + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 85.19 | 89.69 | |
| NSP-SeqKD + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 85.15 | 89.49 | |
| LoRA-SeqKD + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 84.8 | 88.9 | |
| LoRA-SeqKD + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 84.72 | 88.82 | |
| NSP-SeqFT + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 83.39 | 88.47 | |
| LoRA-SeqFT + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 83.39 | 88.47 | |
| SLCA++Backbone=ViT/B-Sup21K2026.01 | 82.53 | 87.07 | |
| HiDe-LoRABackbone=ViT/B-Sup21K2026.01 | 82.24 | 84.16 | |
| CoMABackbone=ViT/B-Sup21K2026.01 | 82.19 | 86.19 | |
| TSVDBackbone=ViT/B-Sup21K2026.01 | 81.54 | — | |
| SeqKD + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 80.88 | 88.05 | |
| NSP-SeqKD + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 80.6 | 87.49 | |
| SLCABackbone=ViT/B-Sup21K2026.01 | 80.24 | 85.78 | |
| RanPACBackbone=ViT/B-Sup21K2026.01 | 79.97 | — | |
| RanProj + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 79.91 | 85.44 | |
| SeqFT + LR-RGDA + HopDCBackbone=ViT/B-Sup21K2026.01 | 78.78 | 86.58 | |
| LoRA-SeqFT + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 76.51 | 83.25 | |
| NSP-SeqFT + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 75.8 | 84.45 | |
| LAE-AdapterBackbone=ViT/B-Sup21K2026.01 | 74.95 | 80.42 | |
| CODA-PromptBackbone=ViT/B-Sup21K2026.01 | 69.72 | 76.03 | |
| SeqFT + LR-RGDABackbone=ViT/B-Sup21K2026.01 | 66.59 | 77.13 |