Image Classification on 11 Fine-Grained Datasets Suite (test)
95Food AccuracyFranca
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FrancaArch=ViT-G/14, Evaluation Protocol=Linear probing2025.07 | 95 | 99.5 | 95.1 | 78.9 | 91.3 | 85.5 | 85 | 97.2 | 97.5 | 99.7 | 91.3 | 92.3 | |
| DINOv2Arch=ViT-G/14, Evaluation Protocol=Linear probing2025.07 | 94.7 | 99.5 | 94.4 | 78.7 | 91.4 | 87.2 | 84.5 | 96.7 | 97.6 | 99.7 | 91.6 | 92.3 | |
| OpenCLIPArch=ViT-G/14, Evaluation Protocol=Linear probing2025.07 | 94.5 | 98.7 | 91 | 84 | 96.1 | 80.2 | 86 | 95.7 | 98.1 | 99.5 | 89.9 | 92.2 | |
| DINOv2Arch=ViT-L/14, Evaluation Protocol=Linear probing, Pre-training setup=distilled from DINOv2-G on LVD-142M2025.07 | 94.3 | 99.3 | 93.4 | 78.7 | 89.9 | 81.5 | 84 | 96.5 | 97.5 | 99.7 | 90.5 | 91.4 | |
| FrancaArch=ViT-L/14, Evaluation Protocol=Linear probing2025.07 | 94.3 | 99.4 | 94.1 | 79.9 | 89.5 | 81.3 | 84.1 | 95.1 | 97.4 | 99.8 | 91.1 | 91.5 | |
| Web-SSLArch=ViT-G/14, Evaluation Protocol=Linear probing2025.07 | 94.1 | 99.4 | 93.1 | 78 | 90.3 | 83.7 | 84.7 | 92.4 | 96.8 | 99.4 | 91.2 | 91.2 | |
| DINOv2Arch=ViT-L/14, Evaluation Protocol=Linear probing, Pre-training setup=reproduced on IN-21K without distillation2025.07 | 93.4 | 99.2 | 93.9 | 78.1 | 89.9 | 81.7 | 82.9 | 95.2 | 87.2 | 99.6 | 90.3 | 90.1 | |
| Web-SSLArch=ViT-L/14, Evaluation Protocol=Linear probing2025.07 | 91 | 98.9 | 90.7 | 77.5 | 88.9 | 80.2 | 83.6 | 93.1 | 95.1 | 98.8 | 90.9 | 89.9 | |
| FrancaArch=ViT-B/14, Evaluation Protocol=Linear probing2025.07 | 90.6 | 98.7 | 90.9 | 77 | 88.7 | 75.2 | 81.7 | 94.1 | 96.2 | 99.7 | 86.2 | 88.9 | |
| MAEArch=ViT-H/14, Evaluation Protocol=Linear probing2025.07 | 78.4 | 96.1 | 83.9 | 63.9 | 56.1 | 63.4 | 75.4 | 89.4 | 95.9 | 92.3 | 57.2 | 77.5 |