Composed Image Retrieval on FashionIQ
57.2Recall@10ORACLE-Q
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T32026.05 | 57.2 | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T32026.05 | 54.6 | — | |
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T22026.05 | 54 | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T22026.05 | 51.2 | — | |
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T12026.05 | 48.7 | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T32026.05 | 48.3 | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T12026.05 | 45.9 | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T22026.05 | 45.7 | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T32026.05 | 44.8 | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T22026.05 | 42.9 | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T12026.05 | 42.1 | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T32026.05 | 41 | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T12026.05 | 40.6 | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T22026.05 | 40.2 | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T12026.05 | 39.3 | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T02026.05 | 38.5 | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T02026.05 | 38.5 | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T02026.05 | 38.5 | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T02026.05 | 38.5 | — | |
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T02026.05 | 38.5 | — | |
| MagicLens-LBackbone=CoCa-L, # Total Params=613M2024.03 | 38 | — | |
| CompoDiffBackbone=CLIP-L, # Total Params=568M2024.03 | 36 | — | |
| PLIBackbone=BLIP-B, # Total Params=224M+2024.03 | 35.9 | — | |
| PLIBackbone=CLIP-L, # Total Params=428M+2024.03 | 35.4 | — | |
| MagicLens-BBackbone=CoCa-B, # Total Params=267M2024.03 | 35.2 | — | |
| MagicLens-LBackbone=CLIP-L, # Total Params=465M2024.03 | 30.7 | — | |
| CIReVLBackbone=CLIP-L, # Total Params=12.5B*2024.03 | 28.6 | — | |
| BLIP-FFApproach=Embedding-based, Fusion=FF2025.03 | 28.5 | — | |
| CIReVLBackbone=CLIP-B, # Total Params=12.3B*2024.03 | 28.3 | — | |
| Context-I2WBackbone=CLIP-L, # Total Params=496M2024.03 | 27.8 | — | |
| MagicLens-BBackbone=CLIP-B, # Total Params=166M2024.03 | 26.3 | — | |
| LinCIRBackbone=CLIP-L, # Total Params=442M2024.03 | 26.3 | — | |
| CLIP-SFApproach=Embedding-based, Fusion=SF2025.03 | 26.2 | — | |
| SEARLEBackbone=CLIP-L, # Total Params=442M2024.03 | 25.6 | — | |
| Pic2WordBackbone=CLIP-L, # Total Params=429M2024.03 | 24.7 | — | |
| SEARLEBackbone=CLIP-B, # Total Params=165M2024.03 | 22.9 | — | |
| PALAVRABackbone=CLIP-B, # Total Params=176M2024.03 | 19.8 | — | |
| GENIUSApproach=Generative, Re-ranking=true2025.03 | 19.2 | — | |
| GENIUSApproach=Generative2025.03 | 13.1 | — | |
| BLIP#Params=224M, Zero-shot=true, Multi-modal Training (Generated Data)=false2024.06 | — | 1.5 | |
| BLIP-FFRetrieval Strategy=Embedding-based, Fusion Strategy=Feature-level Fusion, Re-ranking=false, Retrieval Pool=Task-specific2025.03 | — | 29.2 | |
| BLIP-MM#Params=224M, Zero-shot=true, Multi-modal Training (Generated Data)=true2024.06 | — | 1.23 | |
| CLIP#Params=149M, Zero-shot=true, Multi-modal Training (Generated Data)=false2024.06 | — | 3.56 | |
| CLIP-MM#Params=149M, Zero-shot=true, Multi-modal Training (Generated Data)=true2024.06 | — | 5.55 | |
| CLIP-SFRetrieval Strategy=Embedding-based, Fusion Strategy=Score-level Fusion, Re-ranking=false, Retrieval Pool=Task-specific2025.03 | — | 24.4 | |
| GENIUSRetrieval Strategy=Generative, Fusion Strategy=None, Re-ranking=false, Retrieval Pool=Task-specific2025.03 | — | 13.2 | |
| GENIUSRetrieval Strategy=Generative, Fusion Strategy=None, Re-ranking=true, Retrieval Pool=Task-specific2025.03 | — | 19.3 | |
| Pic2Word#Params=224M, Zero-shot=true, Multi-modal Training (Generated Data)=false2024.06 | — | 8.24 | |
| Pic2Word-MM#Params=224M, Zero-shot=true, Multi-modal Training (Generated Data)=true2024.06 | — | 7.65 | |
| VISTA#Params=196M, Zero-shot=true2024.06 | — | 7.51 |