Text-to-Image Retrieval on MUGE
68.9R@1CN-CLIP
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CN-CLIPEvaluation Setup=Finetuning, Model Size=Huge-size, Backbone=ViT-H/142022.11 | 68.9 | 88.7 | 93.1 | 83.6 | — | |
| CN-CLIPEvaluation Setup=Finetuning, Model Size=Large-size, Backbone=ViT-L/14, Resolution=336px2022.11 | 65.3 | 86.7 | 92.1 | 81.3 | — | |
| CN-CLIPEvaluation Setup=Finetuning, Model Size=Large-size, Backbone=ViT-L/142022.11 | 63.3 | 85.6 | 91.3 | 80.1 | — | |
| CN-CLIPEvaluation Setup=Zero-shot, Model Size=Huge-size, Backbone=ViT-H/142022.11 | 63 | 84.1 | 89.2 | 78.8 | — | |
| R2D2Evaluation Setup=Finetuning, Model Size=Large-size, Backbone=ViT-L/142022.11 | 60.1 | 82.9 | 89.4 | 77.5 | — | |
| CN-CLIPEvaluation Setup=Zero-shot, Model Size=Large-size, Backbone=ViT-L/14, Resolution=336px2022.11 | 59 | 81.4 | 87.8 | 76.1 | — | |
| CN-CLIPEvaluation Setup=Finetuning, Model Size=Base-size, Backbone=ViT-B/162022.11 | 58.4 | 83.6 | 90 | 77.4 | — | |
| CN-CLIPEvaluation Setup=Zero-shot, Model Size=Large-size, Backbone=ViT-L/142022.11 | 56.3 | 79.8 | 86.2 | 74.1 | — | |
| Wukong_ViT-LBackbone=ViT-L, Evaluation Protocol=Fine-tuned2022.02 | 52.7 | 77.9 | 85.6 | 72.1 | — | |
| WukongEvaluation Setup=Finetuning, Model Size=Large-size, Backbone=ViT-L/142022.11 | 52.7 | 77.9 | 85.6 | 72.1 | — | |
| CN-CLIPEvaluation Setup=Zero-shot, Model Size=Base-size, Backbone=ViT-B/162022.11 | 52.1 | 76.7 | 84.4 | 71.1 | — | |
| CLIP_ViT-LBackbone=ViT-L, Evaluation Protocol=Fine-tuned2022.02 | 50.1 | 76.9 | 84.9 | 70.6 | — | |
| R2D2Evaluation Setup=Zero-shot, Model Size=Large-size, Backbone=ViT-L/142022.11 | 49.5 | 75.7 | 83.2 | 69.5 | — | |
| R2D2Backbone=ViT-L, Zero-shot=true2022.05 | 49.5 | 75.7 | 83.2 | — | 69.5 | |
| CN-CLIPEvaluation Setup=Finetuning, Model Size=Tiny-size, Backbone=ResNet-502022.11 | 48.6 | 75.1 | 84 | 69.2 | — | |
| R2D2Evaluation Setup=Finetuning, Model Size=Base-size, Backbone=ViT-B2022.11 | 47.4 | 75.1 | 83.5 | 68.7 | — | |
| CLIP_Swin-LBackbone=Swin-L, Evaluation Protocol=Fine-tuned2022.02 | 45.3 | 72.1 | 81.1 | 66.2 | — | |
| FILIP_Swin-LBackbone=Swin-L, Evaluation Protocol=Fine-tuned2022.02 | 44 | 71.4 | 81.2 | 65.5 | — | |
| Wukong_Swin-LBackbone=Swin-L, Evaluation Protocol=Fine-tuned2022.02 | 43.8 | 71.9 | 81.7 | 65.8 | — | |
| CLIP_ViT-BBackbone=ViT-B, Evaluation Protocol=Fine-tuned2022.02 | 43.5 | 71.7 | 80.6 | 65.3 | — | |
| FILIP_ViT-LBackbone=ViT-L, Evaluation Protocol=Fine-tuned2022.02 | 43.5 | 71.5 | 80.9 | 65.3 | — | |
| CLIPBackbone=ViT-L, Zero-shot=true2022.02 | 43.3 | 69.2 | 78.4 | 63.6 | — | |
| CLIPBackbone=ViT-L, Zero-shot=true2022.05 | 43.3 | 69.2 | 78.4 | — | 63.6 | |
| WukongBackbone=ViT-L, Zero-shot=true2022.02 | 42.7 | 69 | 78 | 63.2 | — | |
| WukongEvaluation Setup=Zero-shot, Model Size=Large-size, Backbone=ViT-L/142022.11 | 42.7 | 69 | 78 | 63.2 | — | |
| WukongBackbone=ViT-L, Zero-shot=true2022.05 | 42.7 | 69 | 78 | — | 63.2 | |
| CN-CLIPEvaluation Setup=Zero-shot, Model Size=Tiny-size, Backbone=ResNet-502022.11 | 42.6 | 68.6 | 77.9 | 63 | — | |
| Wukong_ViT-BBackbone=ViT-B, Evaluation Protocol=Fine-tuned2022.02 | 39.2 | 66.9 | 77.4 | 61.2 | — | |
| WukongEvaluation Setup=Finetuning, Model Size=Base-size, Backbone=ViT-B/322022.11 | 39.2 | 66.9 | 77.4 | 61.2 | — | |
| FILIPBackbone=ViT-L, Zero-shot=true2022.02 | 37.6 | 63.4 | 73.6 | 58.2 | — | |
| FILIPBackbone=ViT-L, Zero-shot=true2022.05 | 37.6 | 63.4 | 73.6 | — | 58.2 | |
| CLIPBackbone=ViT-B, Zero-shot=true2022.02 | 37.3 | 64.2 | 73.9 | 58.5 | — | |
| FILIPBackbone=Swin-L, Zero-shot=true2022.02 | 36.2 | 61.1 | 71.5 | 56.3 | — | |
| CLIPBackbone=Swin-L, Zero-shot=true2022.02 | 35.2 | 62.2 | 73.2 | 56.9 | — | |
| WukongBackbone=Swin-L, Zero-shot=true2022.02 | 34.5 | 60.6 | 71.2 | 55.5 | — | |
| WukongBackbone=ViT-B, Zero-shot=true2022.02 | 33.4 | 59.3 | 69.7 | 54.1 | — | |
| WukongEvaluation Setup=Zero-shot, Model Size=Base-size, Backbone=ViT-B/322022.11 | 33.4 | 59.3 | 69.7 | 54.1 | — | |
| FILIP_ViT-BBackbone=ViT-B, Evaluation Protocol=Fine-tuned2022.02 | 30.6 | 58.2 | 70.2 | 53 | — | |
| FILIPBackbone=ViT-B, Zero-shot=true2022.02 | 22.4 | 46.6 | 58.5 | 42.5 | — | |
| BriVLBackbone=N/A, Zero-shot=true2022.02 | 12.7 | 30.9 | 41.8 | 28.5 | — |