Class-Incremental Learning on Four within-domain datasets average (test)
82.2Last AccuracyJoint-Training
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Joint-TrainingBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 82.2 | — | |
| LoRA-SeqKD + LR-RGDA + HopDCBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 81.32 | 86.82 | |
| SeqKD + LR-RGDA + HopDCBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 81.22 | 85.81 | |
| NSP-SeqKD + LR-RGDA + HopDCBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 79.9 | 84.83 | |
| NSP-SeqFT + LR-RGDA + HopDCBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 79.37 | 84.33 | |
| SeqFT + LR-RGDA + HopDCBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 78.5 | 84.56 | |
| LoRA-SeqFT + LR-RGDA + HopDCBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 78.06 | 84.99 | |
| NSP-SeqKD + LR-RGDABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 75.87 | 83.28 | |
| NSP-SeqFT + LR-RGDABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 74.9 | 82.53 | |
| LoRA-SeqKD + LR-RGDABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 74.84 | 83.93 | |
| SLCA++Backbone=ViT/B-MoCoV3, Random seeds=32026.01 | 74.74 | 80.72 | |
| CoMABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 74.63 | 81.08 | |
| RanPACBackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 73.31 | 80.91 | |
| RanProj + LR-RGDABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 71.91 | 78.52 | |
| SeqKD + LR-RGDABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 71.81 | 81.44 | |
| LoRA-SeqFT + LR-RGDABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 63.02 | 73.23 | |
| SeqFT + LR-RGDABackbone=ViT/B-MoCoV3, Random seeds=32026.01 | 62.61 | 72.95 |