Zero-Shot Classification Suite on CLIP Standard (test val)
45ImageNet AccuracyRILS
Evaluation Results
| Method | Links | ||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RILSEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.01 | 45 | 58.9 | 86.2 | 55.1 | 23.4 | 51.8 | 19.5 | 5.9 | 32.8 | 59.2 | 80.7 | 33.5 | 22.6 | 40.1 | 93.2 | 28.8 | 40.2 | 19.1 | 7.8 | 16.8 | 50 | 42.3 | 17 | — | — | — | — | — | |
| MAE+CLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.01 | 42.3 | 57.8 | 78.2 | 52.4 | 23.9 | 51.6 | 18.1 | 4.6 | 31.5 | 55.8 | 78.4 | 32 | 27.6 | 32.7 | 89.8 | 27 | 39.4 | 22.9 | 7.2 | 14.7 | 49.3 | 39.9 | 0 | — | — | — | — | — | |
| SLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.01 | 41.6 | 56.7 | 73.4 | 43.2 | 22.6 | 51.6 | 17.7 | 4.9 | 32.4 | 52.5 | 79.1 | 33.3 | 29.4 | 33.5 | 89.5 | 17.8 | 36.2 | 17.8 | 6.8 | 23.4 | 49.7 | 38.7 | 2 | — | — | — | — | — | |
| CLIPEvaluation Protocol=Zero-shot, Backbone=ViT-B/16, Training Data=L-20M, Training Epochs=252023.01 | 40.3 | 55.7 | 76 | 46.9 | 24.4 | 50.7 | 17.8 | 4.8 | 31.5 | 53.7 | 78.4 | 31.8 | 26.8 | 37.6 | 89 | 22.7 | 36.9 | 24.1 | 6.8 | 20 | 49.1 | 39.3 | 2 | — | — | — | — | — | |
| Ours-RGB0.3Zero-shot=true, Normalized Training Time (N.T.)=0.64x2024.05 | 36.6 | — | 63.1 | 26.3 | — | — | — | — | — | 57.3 | 72.3 | 25.4 | 9.4 | — | — | — | — | — | — | — | — | 36.8 | — | 8.8 | 39.9 | 45.9 | 24.9 | 31.8 | |
| Ours-EmbeddingZero-shot=true, Normalized Training Time (N.T.)=0.64x2024.05 | 36.3 | — | 70.7 | 32 | — | — | — | — | — | 55.4 | 73.7 | 28.4 | 11 | — | — | — | — | — | — | — | — | 38.2 | — | 8.1 | 39.6 | 47.9 | 25.4 | 30.7 | |
| CLIPZero-shot=true, Normalized Training Time (N.T.)=1.00x2024.05 | 36.1 | — | 57.7 | 25 | — | — | — | — | — | 53.9 | 73.5 | 26 | 11.7 | — | — | — | — | — | — | — | — | 36.1 | — | 8 | 38.4 | 47.6 | 24.9 | 30.7 | |
| Ours-RGB0.5Zero-shot=true, Normalized Training Time (N.T.)=0.54x2024.05 | 36 | — | 63.6 | 27.2 | — | — | — | — | — | 55 | 72.2 | 26.1 | 11.2 | — | — | — | — | — | — | — | — | 36.5 | — | 7.7 | 39.8 | 45.3 | 23.8 | 30.5 | |
| Ours-KmeansZero-shot=true, Normalized Training Time (N.T.)=0.70x2024.05 | 35.5 | — | 61.4 | 25.3 | — | — | — | — | — | 52.1 | 69.9 | 25.6 | 9.7 | — | — | — | — | — | — | — | — | 35 | — | 7.2 | 38.3 | 43 | 22.2 | 30.1 | |
| FLIP AttnZero-shot=true, Normalized Training Time (N.T.)=1.06x2024.05 | 35.2 | — | 61.6 | 27.1 | — | — | — | — | — | 51.2 | 73.5 | 25.7 | 9.4 | — | — | — | — | — | — | — | — | 35.8 | — | 8.1 | 39.4 | 45.1 | 23.7 | 30.1 | |
| FLIPZero-shot=true, Normalized Training Time (N.T.)=0.53x2024.05 | 34.4 | — | 52.8 | 24.5 | — | — | — | — | — | 46 | 70.4 | 25.3 | 10.4 | — | — | — | — | — | — | — | — | 33.5 | — | 7.1 | 39.5 | 41.4 | 20.1 | 29.5 |