Base-to-novel Classification on Base-to-novel Evaluation Suite (ViT-B/32 backbone, base classes)
82.6Average AccuracyCLIP-LoRA† + ADK
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CLIP-LoRA† + ADKBackbone=ViT-B/322025.12 | 82.6 | 72.6 | 80.6 | 42.8 | 95.6 | 80.1 | 83.8 | 91.6 | 97.4 | 97.7 | 80.6 | 85.8 | |
| 2SFS† + ADKBackbone=ViT-B/322025.12 | 82.5 | 73 | 81.2 | 40.7 | 95.3 | 78.9 | 85.1 | 92 | 97.3 | 97.8 | 80.7 | 85 | |
| CLIP-LoRA†Backbone=ViT-B/322025.12 | 82.3 | 72.1 | 80.3 | 41.3 | 96.2 | 79 | 83.8 | 91.7 | 97.4 | 97.7 | 80.4 | 84.9 | |
| 2SFS†Backbone=ViT-B/322025.12 | 82.1 | 72.7 | 81 | 39.5 | 94.9 | 78.7 | 84.8 | 91.6 | 97.2 | 97.7 | 79.8 | 85 | |
| Rep-Adapter† + ADKBackbone=ViT-B/322025.12 | 80.8 | 72.2 | 80.3 | 37.4 | 93.2 | 77.5 | 82.5 | 88.8 | 96.8 | 97.3 | 78.2 | 84.3 | |
| Rep-Adapter†Backbone=ViT-B/322025.12 | 80.3 | 71.7 | 79.4 | 37.4 | 93.1 | 76.9 | 81.8 | 87.8 | 96.7 | 97.2 | 78.1 | 83.7 | |
| MMABackbone=ViT-B/322025.12 | 78.7 | 72.5 | 80.3 | 31.8 | 71.8 | 73.7 | 85.8 | 93.8 | 95.5 | 97.2 | 79.5 | 83.8 | |
| CLIPBackbone=ViT-B/322025.12 | 67.3 | 67.5 | 69.8 | 21.3 | 55.1 | 60.7 | 85.3 | 90.6 | 72.4 | 94.1 | 54.2 | 69.1 |