Few-shot Classification on 6 Datasets (ImageNet, UCF101, Flowers102, Food101, OxfordPets, Caltech101)
88.8Adaptation ScoreCLIP-LoRA + FMA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CLIP-LoRA + FMABackbone=ViT-B/16, Step size (h)=0.12025.10 | 88.8 | 80.6 | 84.5 | |
| CLIP-LoRABackbone=ViT-B/162025.10 | 88.2 | 80.5 | 84.2 | |
| CoOp + FMABackbone=ViT-B/16, Step size (h)=0.12025.10 | 87.9 | 75.7 | 81.3 | |
| CLIP + FMABackbone=ViT-B/16, Step size (h)=0.12025.10 | 87.6 | 81 | 84.2 | |
| CLIP-Adapter + FMABackbone=ViT-B/16, Step size (h)=0.12025.10 | 87.4 | 80 | 83.5 | |
| CoCoOp + FMABackbone=ViT-B/16, Step size (h)=0.12025.10 | 87.3 | 82.2 | 84.7 | |
| CoOpBackbone=ViT-B/162025.10 | 87.1 | 75.8 | 81.1 | |
| CLIP-AdapterBackbone=ViT-B/162025.10 | 86.6 | 79.8 | 83.1 | |
| CoCoOpBackbone=ViT-B/162025.10 | 85 | 82.1 | 83.5 | |
| CLIPBackbone=ViT-B/162025.10 | 78.8 | 81.2 | 79.6 |