Text-to-Image Retrieval on COCO 2017 (val)
48.5Recall@1SmartCLIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SmartCLIPModel Scale=L/142025.07 | 48.5 | 73.1 | 81.7 | |
| Long-CLIPModel Scale=L/142025.07 | 46.3 | 70.8 | 79.8 | |
| SmartCLIPModel Scale=B/162025.07 | 42.4 | 68.2 | 77.8 | |
| Long-CLIPModel Scale=B/162025.07 | 40.4 | 65.8 | 75.2 | |
| CLIP-RefineBackbone=ViT-B/32, Zero-shot=true2025.04 | 37.64 | 63.54 | 74.42 | |
| m²-mixBackbone=ViT-B/32, Zero-shot=true2025.04 | 36.28 | 62.18 | 73.08 | |
| HyCDBackbone=ViT-B/32, Zero-shot=true2025.04 | 36.04 | 62.28 | 73.14 | |
| CLIPModel Scale=L/142025.07 | 35.4 | 60.1 | 70.2 | |
| ContrastiveBackbone=ViT-B/32, Zero-shot=true2025.04 | 34.88 | 61.5 | 72.1 | |
| HyCD + LalignBackbone=ViT-B/32, Zero-shot=true2025.04 | 33.92 | 61.18 | 72.06 | |
| CLIPModel Scale=B/162025.07 | 32.7 | 57.7 | 68.2 | |
| Self-KDBackbone=ViT-B/32, Zero-shot=true2025.04 | 31.04 | 55.58 | 65.9 | |
| Pre-trained (CLIP)Backbone=ViT-B/32, Zero-shot=true2025.04 | 30.56 | 54.92 | 65.26 | |
| Direct Fine-tuningModel Scale=L/142025.07 | 23.1 | 45.1 | 55.9 | |
| Direct Fine-tuningModel Scale=B/162025.07 | 21.8 | 43.4 | 54.5 |