Base-to-novel Classification on Base-to-novel Evaluation Suite (Novel Classes)
73.9Avg Accuracy2SFS† + ADK
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2SFS† + ADKBackbone=ViT-B/322025.12 | 73.9 | 67.2 | 78.4 | 34.5 | 71.5 | 70.5 | 87.5 | 96.6 | 74.8 | 94.4 | 61.6 | 75.8 | |
| CLIP-LoRA† + ADKBackbone=ViT-B/322025.12 | 71.9 | 67.4 | 78.6 | 31.1 | 65.9 | 67.9 | 85.4 | 94.7 | 70 | 93.9 | 61.3 | 74.5 | |
| CLIPBackbone=ViT-B/322025.12 | 71.7 | 64.1 | 73 | 29.3 | 69.8 | 69.7 | 86.9 | 96.9 | 73.7 | 94 | 58.2 | 73 | |
| 2SFS†Backbone=ViT-B/322025.12 | 71.4 | 66.7 | 77.7 | 30.3 | 63.8 | 70.2 | 87 | 95.7 | 70.2 | 93.7 | 55.1 | 74.7 | |
| MMABackbone=ViT-B/322025.12 | 71 | 65.8 | 76.6 | 28.7 | 63 | 69.3 | 87.1 | 96.3 | 71.6 | 92.6 | 57 | 73.5 | |
| CLIP-LoRA†Backbone=ViT-B/322025.12 | 70 | 66.7 | 77.8 | 25.8 | 60.3 | 68.5 | 85.8 | 95.8 | 67.9 | 92.9 | 56 | 72.3 | |
| Rep-Adapter† + ADKBackbone=ViT-B/322025.12 | 69.4 | 66.9 | 76.5 | 27.7 | 60.6 | 65.3 | 85.4 | 93.9 | 65.3 | 93.4 | 55.9 | 72.5 | |
| Rep-Adapter†Backbone=ViT-B/322025.12 | 67.5 | 66 | 74.7 | 23.8 | 58.1 | 64.5 | 85.1 | 94.6 | 62 | 92.8 | 52.8 | 68 |