Classification on Downstream Classification Suite
96.6CIFAR10 AccuracyFFF-ViT-B/32
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FFF-ViT-B/32Backbone=ViT-B/32, Pre-train dataset=Open30M, Evaluation protocol=Linear probe2024.05 | 96.6 | 84.1 | 83.8 | 87.4 | 95.7 | 97.3 | 74.1 | 75.5 | 99.9 | 38.7 | 83.3 | |
| DeCLIP-VIT-B/32Backbone=ViT-B/32, Pre-train dataset=Open30M, Evaluation protocol=Linear probe2024.05 | 93.1 | 76.9 | 82 | 82.7 | 96 | 74.9 | 59.8 | 74.5 | 92.6 | 32.7 | 76.5 | |
| HiCLIP-ViT-B/32Backbone=ViT-B/32, Pre-train dataset=Open30M, Evaluation protocol=Linear probe2024.05 | 92.8 | 75.8 | 80.5 | 81.3 | 94.4 | 73.6 | 59.4 | 72.2 | 90.3 | 33.6 | 75.4 | |
| HiDeCLIP-ViT-B/32Backbone=ViT-B/32, Pre-train dataset=Open30M, Evaluation protocol=Linear probe2024.05 | 92.7 | 75.6 | 82.9 | 83.3 | 95.7 | 75.6 | 62.8 | 74.5 | 92 | 35.8 | 77.1 | |
| CLIP-ViT-B/32Backbone=ViT-B/32, Pre-train dataset=Open30M, Evaluation protocol=Linear probe2024.05 | 92 | 74.7 | 78.8 | 80.7 | 93.7 | 72.6 | 55.9 | 71.4 | 88.6 | 29.7 | 73.8 |