Image Classification on VTAB+ 1.0 (test)
38.9Top-1 AccuracyBaseline
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BaselineModel Family=ViT Family, Model=ViT-L/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 38.9 | 25.33 | 43.25 | 16.28 | 35.52 | |
| CoM-PTModel Family=ViT Family, Model=ViT-L/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 38.89 | 6.62 | 15.24 | 4.52 | 14.97 | |
| BaselineModel Family=ViT Family, Model=ViT-B/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 36.98 | 8.79 | 17.92 | 7.94 | 19.24 | |
| CoM-PTModel Family=ViT Family, Model=ViT-B/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 36.89 | 2.82 | 8.62 | 2.75 | 10.45 | |
| BaselineModel Family=ViT Family, Model=ViT-S/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 35.35 | 3.61 | 3.61 | 5.1 | 5.1 | |
| CoM-PTModel Family=ViT Family, Model=ViT-S/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 35.35 | 3.61 | 3.61 | 5.1 | 5.1 | |
| CoM-PTModel Family=ViT Family, Model=ViT-M/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 35.22 | 2.19 | 5.8 | 2.6 | 7.7 | |
| BaselineModel Family=ViT Family, Model=ViT-M/16, Pre-training Dataset=Merged-15M, Evaluation Protocol=Zero-shot2026.04 | 34.97 | 5.52 | 9.13 | 6.2 | 11.3 | |
| CoM-PTModel Family=Swin Family, Model=Swin-L, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 32.68 | 0.84 | 2.73 | 0.73 | 4.21 | |
| CoM-PTModel Family=Swin Family, Model=Swin-B, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 32.53 | 0.52 | 1.89 | 0.58 | 3.48 | |
| BaselineModel Family=Swin Family, Model=Swin-L, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 31.8 | 7.86 | 14.25 | 6.76 | 16 | |
| CoM-PTModel Family=Swin Family, Model=Swin-S, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 31.35 | 0.43 | 1.37 | 0.68 | 2.9 | |
| BaselineModel Family=Swin Family, Model=Swin-B, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 31.17 | 3.48 | 6.39 | 3.73 | 9.24 | |
| CoM-PTModel Family=ViT Family, Model=ViT-L/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 30.78 | 1.41 | 2.94 | 1.02 | 3.64 | |
| BaselineModel Family=Swin Family, Model=Swin-S, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 30.55 | 1.97 | 2.91 | 3.29 | 5.51 | |
| BaselineModel Family=ViT Family, Model=ViT-L/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 30.39 | 10.79 | 16.7 | 7.56 | 14.95 | |
| BaselineModel Family=ViT Family, Model=ViT-B/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 30.08 | 3.75 | 5.91 | 3.45 | 7.39 | |
| CoM-PTModel Family=ViT Family, Model=ViT-B/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 29.89 | 0.59 | 1.53 | 0.54 | 2.62 | |
| BaselineModel Family=ViT Family, Model=ViT-S/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 28.84 | 1.54 | 2.16 | 2.4 | 3.94 | |
| BaselineModel Family=Swin Family, Model=Swin-T, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 28.72 | 0.94 | 0.94 | 2.22 | 2.22 | |
| CoM-PTModel Family=Swin Family, Model=Swin-T, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 28.72 | 0.94 | 0.94 | 2.22 | 2.22 | |
| CoM-PTModel Family=ViT Family, Model=ViT-S/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 28.61 | 0.32 | 0.94 | 0.52 | 2.06 | |
| BaselineModel Family=ViT Family, Model=ViT-T/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 25.61 | 0.62 | 0.62 | 1.54 | 1.54 | |
| CoM-PTModel Family=ViT Family, Model=ViT-T/16, Pre-training Dataset=CC3M, Evaluation Protocol=Zero-shot2026.04 | 25.61 | 0.62 | 0.62 | 1.54 | 1.54 |