Continual Learning on CUB-200 Split
90.9Final Avg AccuracyNoRGa
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| NoRGaBackbone=ViT-B/16, Pre-training=Sup-21K2024.05 | 90.9 | — | — | |
| RanPACBackbone=ViT-B/16, Pre-training=Sup-21K2024.05 | 90.3 | — | — | |
| ADAM + AdapterBackbone=ViT-B/16, Pre-training=Sup-21K2024.05 | 85.84 | — | — | |
| ADAM + SSFBackbone=ViT-B/16, Pre-training=Sup-21K2024.05 | 85.67 | — | — | |
| ADAM + VPT-DBackbone=ViT-B/16, Pre-training=Sup-21K2024.05 | 85.28 | — | — | |
| TreeLoRATIME (s)=86 ± 32025.06 | 73.66 | — | 4.87 | |
| HiDeLoRATIME (s)=194 ± 32025.06 | 73.48 | — | 5.38 | |
| HiDePromptTIME (s)=307 ± 12025.06 | 63.86 | — | 5.57 | |
| GEMTIME (s)=7233 ± 182025.06 | 61.28 | — | 2.64 | |
| L2PTIME (s)=333 ± 52025.06 | 39.72 | — | 7.98 | |
| Adaptive TrainingTraining Mode=Adaptive2026.02 | 39.01 | 20.76 | — | |
| Initialized TrainingTraining Mode=Initialized2026.02 | 37.98 | 21.28 | — | |
| DualPromptTIME (s)=315 ± 42025.06 | 37.52 | — | 11.91 | |
| Frozen TrainingTraining Mode=Frozen2026.02 | 37.15 | 21.18 | — | |
| EWCTIME (s)=1702 ± 92025.06 | 36.34 | — | 6.65 | |
| SeqLoRATIME (s)=194 ± 112025.06 | 9.1 | — | 83.45 |