Image Classification on 13 Transfer Learning Datasets (val)
48.6Aircraft AccuracyDINO + DILEMMA
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DINO + DILEMMABackbone=ViT-S/16, Evaluation Protocol=Linear probing, Training Duration=same as baseline2022.04 | 48.6 | 89.66 | 50.85 | 93.39 | 78.85 | 68.3 | 95.38 | 75.48 | 53.71 | 86.54 | 97.41 | 71.47 | 64 | 74.82 | |
| DINO + DILEMMABackbone=ViT-S/16, Evaluation Protocol=Linear probing2022.04 | 46.95 | 89.58 | 49.04 | 92.46 | 77.22 | 67.5 | 95.2 | 74.9 | 52.51 | 85.91 | 97.06 | 70.01 | 62.87 | 73.93 | |
| DINO + SparsityBackbone=ViT-S/16, Evaluation Protocol=Linear probing, Sparsity=True2022.04 | 46.83 | 89.16 | 48.45 | 92.47 | 77 | 68.4 | 94.7 | 74.97 | 51.32 | 85.45 | 96.85 | 66.65 | 61.9 | 73.4 | |
| MoCoV3 + DILEMMABackbone=ViT-S/16, Evaluation Protocol=Linear probing, Training Duration=same as baseline2022.04 | 46.02 | 90.29 | 43.44 | 94.2 | 80.05 | 65.37 | 94.47 | 74.1 | 44.13 | 88.53 | 96.75 | 66.3 | 64.9 | 72.97 | |
| DINOBackbone=ViT-S/16, Evaluation Protocol=Linear probing2022.04 | 45.66 | 88.3 | 47.07 | 90.61 | 74.06 | 66.22 | 94.54 | 73 | 47.36 | 85.83 | 96.62 | 53.74 | 59.9 | 70.99 | |
| MoCoV3 + SparsityBackbone=ViT-S/16, Evaluation Protocol=Linear probing, Sparsity=True, Training Duration=same as baseline2022.04 | 44.64 | 89.89 | 40.21 | 93.53 | 78.88 | 65.48 | 94.21 | 72.96 | 42.84 | 88.93 | 96.9 | 64.52 | 63.24 | 72.02 | |
| MoCoV3 + DILEMMABackbone=ViT-S/16, Evaluation Protocol=Linear probing2022.04 | 44.43 | 89.55 | 42.32 | 93.03 | 77.56 | 64.47 | 93.41 | 71.94 | 43.76 | 85.88 | 96.08 | 65.41 | 63.76 | 71.66 | |
| MoCoV3 + SparsityBackbone=ViT-S/16, Evaluation Protocol=Linear probing, Sparsity=True2022.04 | 43.29 | 89.25 | 40.14 | 92.28 | 77.3 | 65.05 | 93.25 | 71.39 | 42.38 | 86.35 | 96.09 | 64.6 | 62.29 | 71.05 | |
| MoCoV3Backbone=ViT-S/16, Evaluation Protocol=Linear probing, Positional Embeddings=True2022.04 | 38.7 | 87.35 | 28.72 | 91.97 | 75.09 | 64.63 | 91.67 | 67.97 | 33.3 | 84.63 | 95.76 | 64.56 | 56.41 | 67.75 | |
| MoCoV3Backbone=ViT-S/16, Evaluation Protocol=Linear probing, Positional Embeddings=False2022.04 | 16.29 | 60.79 | 5.88 | 31.36 | 13.96 | 50.59 | 60.51 | 31.64 | 18.12 | 53.34 | 70.61 | 20.02 | 15.25 | 34.49 |