Fine-grained classification on Oxford Flowers 102
99.69AccuracyDINOv3-B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DINOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 99.69 | — | |
| DINOv2-Reg-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 99.67 | — | |
| DINOv2-BEvaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 99.59 | — | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=642026.05 | 99.35 | — | |
| VECA-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 256, Core budget (C)=82026.05 | 98.98 | — | |
| DFN2B-CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 98.7 | — | |
| C-RADIOv3-BEvaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 98.52 | — | |
| SigLIP2-B/16-256Evaluation protocol=frozen-backbone linear probing, Crop size=256 x 2562026.05 | 97.98 | — | |
| OpenCLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 96.5 | — | |
| CLIP-B/16Evaluation protocol=frozen-backbone linear probing, Crop size=224 x 2242026.05 | 96.36 | — | |
| TPTPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.04 | 76.9 | 0 | |
| C-TPTPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.04 | 76.3 | 0 | |
| CLIPPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.04 | 76.2 | 0 | |
| R-TPTPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.04 | 76.2 | 55.6 | |
| MTAPre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.04 | 76.1 | 44.2 | |
| EnsemblePre-trained Backbone=CLIP-ViT-L/14, Perturbation budget (epsilon)=4.02025.04 | 75 | 48.6 | |
| ADAPTBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Transductive2025.08 | 74.14 | — | |
| TransCLIPBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Transductive2025.08 | 72.2 | — | |
| ADAPTBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 70.04 | — | |
| TDABackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 68.74 | — | |
| DotaBackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 68.53 | — | |
| TPSBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 68.2 | — | |
| DMNBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 67.93 | — | |
| StatABackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Transductive2025.08 | 67.7 | — | |
| DPEBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 67.6 | — | |
| BCABackbone=CLIP-RN50, BP-free=true, Evaluation Protocol=Online2025.08 | 66.3 | — | |
| C-TPTBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 65.2 | — | |
| CLIPPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.04 | 64 | 0 | |
| DiffTPTBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 63.53 | — | |
| TPTBackbone=CLIP-RN50, BP-free=false, Evaluation Protocol=Online2025.08 | 62.69 | — | |
| R-TPTPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.04 | 62.6 | 37.6 | |
| CLIPBackbone=CLIP-RN50, BP-free=N/A, Evaluation Protocol=Online2025.08 | 61.75 | — | |
| CLIPPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.04 | 61.7 | 0 | |
| R-TPTPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.04 | 60.6 | 51.9 | |
| APT+TeCoAPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.04 | 42.7 | 42.6 | |
| APT+TeCoAPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.04 | 42.5 | 40.4 | |
| FAREPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.04 | 37 | 34 | |
| TeCoAPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.04 | 33.5 | 33.4 | |
| TeCoAPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.04 | 30.8 | 28.9 | |
| APTPre-trained Backbone=ResNet50, Perturbation budget (epsilon)=1.02025.04 | 2.6 | 1.1 | |
| APTPre-trained Backbone=ViT-B/32, Perturbation budget (epsilon)=4.02025.04 | 0.9 | 0.2 |