Multi-Label Incremental Learning on MS COCO B40-C10 protocol official (val)
83.1Last mAPJoint
Evaluation results
| Method | Links | ||
|---|---|---|---|
| JointBackbone=ViT, PEFT=-, Buffer size=-2026.01 | 83.1 | — | |
| JointBackbone=CNN, PEFT=-, Buffer size=-2026.01 | 81.8 | — | |
| P2L-CA+Backbone=ViT, PEFT=✓, Buffer size=0/class2026.01 | 81.4 | 84.1 | |
| P2L-CABackbone=ViT, PEFT=✓, Buffer size=0/class2026.01 | 78.6 | 81.4 | |
| HCPBackbone=CNN, PEFT=×, Buffer size=0/class2026.01 | 75.3 | 78.9 | |
| KRT-RBackbone=CNN, PEFT=×, Buffer size=20/class2026.01 | 75.2 | 78.3 | |
| CSCBackbone=CNN, PEFT=×, Buffer size=0/class2026.01 | 75 | 78.2 | |
| KRTBackbone=CNN, PEFT=×, Buffer size=0/class2026.01 | 74 | 77.8 | |
| KRT*Backbone=ViT, PEFT=×, Buffer size=0/class2026.01 | 73.6 | 77.2 | |
| CODABackbone=ViT, PEFT=✓, Buffer size=0/class2026.01 | 72.4 | 75.3 | |
| DualPromptBackbone=ViT, PEFT=✓, Buffer size=0/class2026.01 | 71.9 | 74.5 | |
| L2PBackbone=ViT, PEFT=✓, Buffer size=0/class2026.01 | 70.4 | — | |
| AGCNBackbone=CNN, PEFT=×, Buffer size=0/class2026.01 | 69.1 | 73.9 | |
| DER++Backbone=CNN, PEFT=×, Buffer size=20/class2026.01 | 66.3 | 73.6 | |
| PODNetBackbone=CNN, PEFT=×, Buffer size=20/class2026.01 | 64.2 | 71 | |
| ERBackbone=CNN, PEFT=×, Buffer size=20/class2026.01 | 61.6 | 68.9 | |
| LWFBackbone=CNN, PEFT=×, Buffer size=0/class2026.01 | 29.9 | 48.6 | |
| oEWCBackbone=CNN, PEFT=×, Buffer size=0/class2026.01 | 27.3 | 44.8 | |
| Fine-TuningBackbone=-, PEFT=-, Buffer size=-2026.01 | 17 | 35.1 |