Text-to-Image Retrieval on Fashion200K
34.9Recall@5CLIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CLIPvariant=oracle2026.04 | 34.9 | — | — | |
| MOON2.0mode=Zero-shot2025.11 | 25.25 | 31.39 | 13.05 | |
| SemiCLIP2026.04 | 23.4 | — | — | |
| ToMA-domain2026.04 | 23.3 | — | — | |
| ToMA2026.04 | 23.2 | — | — | |
| MOONmode=Zero-shot2025.11 | 22.89 | 27.73 | 10.82 | |
| GMEmode=Zero-shot2025.11 | 20.31 | 26.06 | 8.13 | |
| BGE-VL-Largemode=Zero-shot2025.11 | 18.62 | 23.71 | 6.75 | |
| S-CLIP2026.04 | 17.4 | — | — | |
| Soft-PL2026.04 | 15.9 | — | — | |
| MM-Embedmode=Zero-shot2025.11 | 15.71 | 22.25 | 5.97 | |
| Hard-PL2026.04 | 14.5 | — | — | |
| CLIPvariant=fine-tuned2026.04 | 13.5 | — | — | |
| CLIPvariant=original2026.04 | 12.3 | — | — | |
| FashionCLIPmode=Zero-shot2025.11 | 11.67 | 15.14 | 4.92 | |
| CASLIE-Smode=Zero-shot2025.11 | 11.25 | 14.12 | 4.71 | |
| SigLIP2mode=Zero-shot2025.11 | 10.03 | 13.9 | 4.19 | |
| Qwen2.5-VL-3Bmode=Zero-shot2025.11 | 1.06 | 1.5 | 0.45 | |
| InternVL3-2Bmode=Zero-shot2025.11 | 0.27 | 0.55 | 0.09 | |
| BLIP-FFRetrieval Strategy=Embedding-based, Fusion Strategy=Feature-level Fusion, Re-ranking=false, Retrieval Pool=Task-specific2025.03 | — | 26.1 | — | |
| BLIP-FFRetrieval Pool=Global, Approach=Embedding-based, Fusion=Feature-level2025.03 | — | 25.4 | — | |
| CLIPversion=original2026.04 | — | — | 4.5 | |
| CLIPversion=fine-tuned2026.04 | — | — | 4.2 | |
| CLIP-SFRetrieval Strategy=Embedding-based, Fusion Strategy=Score-level Fusion, Re-ranking=false, Retrieval Pool=Task-specific2025.03 | — | 18 | — | |
| CLIP-SFRetrieval Pool=Global, Approach=Embedding-based, Fusion=Score-level2025.03 | — | 17.8 | — | |
| GENIUSRetrieval Strategy=Generative, Fusion Strategy=None, Re-ranking=false, Retrieval Pool=Task-specific2025.03 | — | 13.7 | — | |
| GENIUSRetrieval Strategy=Generative, Fusion Strategy=None, Re-ranking=true, Retrieval Pool=Task-specific2025.03 | — | 16.2 | — | |
| GENIUSRetrieval Pool=Global, Approach=Generative2025.03 | — | 13.7 | — | |
| GENIUSRetrieval Pool=Global, Approach=Generative, Re-ranking=true2025.03 | — | 16.2 | — | |
| Hard-PL2026.04 | — | — | 5.1 | |
| S-CLIP2026.04 | — | — | 6.1 | |
| SemiCLIP2026.04 | — | — | 9.1 | |
| Soft-PL2026.04 | — | — | 5.3 | |
| ToMA2026.04 | — | — | 9 | |
| ToMA-domain2026.04 | — | — | 8.9 |