Image Classification on MetaShift Cat vs. Dog 18 (test)
92.4Test AccuracyOpenCLIP ZS prompt (full)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OpenCLIP ZS prompt (full)Backbone=ViT-L-142026.06 | 92.4 | 88.8 | |
| Full onlyBackbone=ViT-L (300M), Image size=518x5182026.06 | 83.2 | 82.4 | |
| Full onlyBackbone=ViT-L (300M)2026.06 | 80 | 79.2 | |
| Full onlyBackbone=ViT-G (1.1B)2026.06 | 78.6 | 76.8 | |
| A^2 cross-modelBackbone=ViT-S × ViT-G2026.06 | 77.3 | 76.6 | |
| Center CropBackbone=ViT-G (1.1B)2026.06 | 76.9 | 74.8 | |
| Attending on Attention (A^2)Backbone=ViT-G (1.1B)2026.06 | 76.1 | 73.7 | |
| DFRBackbone=ViT-G (1.1B), Requires group labels=true2026.06 | 76 | 74.2 | |
| Full onlyBackbone=ViT-B (86M), Image size=518x5182026.06 | 75.3 | 73.9 | |
| Attending on Attention (A^2)Backbone=ViT-B (86M)2026.06 | 73.9 | 71.8 | |
| DFRBackbone=ViT-L (300M), Requires group labels=true2026.06 | 73.6 | 70.4 | |
| Attending on Attention (A^2)Backbone=ViT-L (300M)2026.06 | 72.5 | 70.6 | |
| Attending on Attention (A^2)Backbone=ViT-S (21M)2026.06 | 69.8 | 68.3 | |
| DFR + A^2Backbone=ViT-S (21M), Requires group labels=true2026.06 | 69.4 | 67.7 | |
| DFRBackbone=ViT-B (86M), Requires group labels=true2026.06 | 68.6 | 65.9 | |
| Full onlyBackbone=ViT-B (86M)2026.06 | 68.5 | 68.2 | |
| Full onlyBackbone=ViT-S (21M), Image size=518x5182026.06 | 65.2 | 60.9 | |
| Center CropBackbone=ViT-S (21M)2026.06 | 64 | 61.3 | |
| Full onlyBackbone=ViT-S (21M)2026.06 | 59.8 | 56 | |
| Scaling on ScalesBackbone=ViT-S (21M), scales=[1, 2], Image size=224x2242026.06 | 59.7 | 56.9 | |
| DFRBackbone=ViT-S (21M), Requires group labels=true2026.06 | 58.7 | 54.5 | |
| Random CropsBackbone=ViT-S (21M)2026.06 | 44 | 40.6 |