Class-Incremental Learning on ImageNet-R 10 tasks
83.2Accuracy (10 Tasks)TRM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TRMBackbone=CLIP ViT-B/16, Method Category=Model Merging2026.05 | 83.2 | — | |
| LwFMemory=Theta(1)2025.09 | 82.97 | — | |
| CMMMemory=Theta(1)2025.09 | 82.69 | — | |
| EWCMemory=Theta(1)2025.09 | 82.42 | — | |
| MaxAbsMemory=Theta(N)2025.09 | 82.33 | — | |
| TIESMemory=Theta(N)2025.09 | 82.27 | — | |
| RandMixMemory=Theta(N)2025.09 | 81.88 | — | |
| AvgMemory=Theta(N)2025.09 | 81.87 | — | |
| Joint2025.11 | 81.14 | — | |
| InfLoRA2024.03 | 80.82 | 75.65 | |
| MagMaxYear=2024, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.05 | 80.1 | — | |
| PMYear=2025, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.05 | 79.9 | — | |
| BECAMEYear=2025, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.05 | 79.8 | — | |
| RAPFYear=2024, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.05 | 79.5 | — | |
| TIESYear=2023, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.05 | 79.5 | — | |
| CLG-CBMYear=2025, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.05 | 78.8 | — | |
| Model StockYear=2024, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.05 | 77.1 | — | |
| HiDe-Prompt2024.03 | 76.6 | 75.06 | |
| CODAPromptYear=2023, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.05 | 76 | — | |
| DualPromptYear=2023, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.05 | 75.9 | — | |
| L2PYear=2022, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.05 | 75.7 | — | |
| SeqLoRA2024.03 | 74.78 | 64.32 | |
| MIRA2025.11 | 73.08 | — | |
| C-LoRA2025.11 | 71.89 | — | |
| LAE2025.11 | 71.7 | — | |
| CODA-P2025.11 | 71.47 | — | |
| EWCYear=2017, Backbone=CLIP ViT-B/16, Method Category=Conventional2026.05 | 66.7 | — | |
| DualPrompt2025.11 | 65.41 | — | |
| L2P2025.11 | 62.54 | — | |
| LwFYear=2017, Backbone=CLIP ViT-B/16, Method Category=Conventional2026.05 | 60.9 | — | |
| Sequential2025.11 | 46.07 | — |