Composed Image Retrieval on FashionIQ (val)
58.16Average Recall@10FashionBLIP-2 + FIRE-CIR
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FashionBLIP-2 + FIRE-CIRbase_model=FashionBLIP-2, enhancement=FIRE-CIR2026.04 | 58.16 | 58.68 | 75.76 | 54.09 | 75.41 | 61.7 | 80.88 | 77.35 | — | — | 67.75 | |
| WRF4CIRBackbone=BLIP-2 ViT-G, Leverage additional data=false2026.04 | 57.35 | 58.94 | 76.21 | 51.71 | 75.41 | 61.4 | 80.37 | 77.33 | — | — | 67.34 | |
| CSMCIR2026.01 | 57.07 | 57.7 | 75.76 | 52.45 | 74.81 | 61.14 | 80.98 | 77.27 | — | — | 67.17 | |
| ReCALLStage=Rrefine2026.02 | 57.04 | 58.49 | 76.59 | 51.81 | 73.48 | 60.83 | 79.19 | 76.42 | — | — | — | |
| DetailFusion2026.04 | 56.89 | 58.12 | 75.95 | 51.34 | 74.05 | 61.22 | 80.09 | 76.7 | — | — | 66.79 | |
| TMCIRMode=Ours2025.04 | 56.57 | 59.12 | 76.34 | 50.67 | 73.86 | 59.93 | 79.46 | 76.55 | — | — | 66.56 | |
| SPRC+SPNBackbone=BLIP-2 ViT-G, Leverage additional data=true2026.04 | 56.37 | 57.7 | 75.27 | 50.57 | 74.12 | 60.84 | 79.96 | 76.45 | — | — | 66.41 | |
| CIR-LVLMVenue=AAAI’252026.02 | 56.21 | 58.59 | 75.86 | 50.42 | 73.6 | 59.61 | 78.99 | 76.14 | — | — | — | |
| CIR-LVLMBackbone=Qwen-7B, Leverage additional data=false2026.04 | 56.21 | 58.59 | 75.86 | 50.42 | 73.57 | 59.61 | 78.99 | 76.14 | — | — | 66.17 | |
| Air-KnowNoise Ratio=0%, Backbone=BLIP-22026.04 | 55.75 | 56.76 | 75.05 | 51.03 | 73.08 | 59.47 | 79.02 | 75.72 | — | — | 65.73 | |
| FashionBLIP-22026.04 | 55.73 | 56.82 | 75.27 | 51.56 | 73.43 | 58.8 | 79.65 | 76.12 | — | — | 65.92 | |
| ConeSepNoise=0%2026.04 | 55.58 | 56.98 | 75.22 | 50.96 | 73.02 | 58.8 | 79.4 | 75.88 | — | — | 65.73 | |
| RUNCBackbone=BLIP-2 ViT-G, Leverage additional data=false2026.04 | 55.52 | 57.26 | 75.32 | 48.93 | 73.53 | 60.38 | 79.86 | 76.23 | — | — | 65.88 | |
| HABITNoise Ratio=0%2026.04 | 55.38 | 56.62 | 74.68 | 49.99 | 72.38 | 59.51 | 78.53 | 75.2 | — | — | 65.29 | |
| HABITNoise=0%2026.04 | 55.38 | 56.62 | 74.68 | 49.99 | 72.38 | 59.51 | 78.53 | 75.2 | — | — | 65.29 | |
| SPRC + FIRE-CIRbase_model=SPRC, enhancement=FIRE-CIR2026.04 | 55.35 | 56.23 | 75.12 | 50.32 | 73.77 | 59.51 | 79.4 | 76.1 | — | — | 65.73 | |
| INTENTNoise=0%2026.04 | 55.31 | 56.32 | 74.93 | 50.32 | 72.1 | 59.28 | 78.45 | 75.16 | — | — | 65.24 | |
| INTENTNoise Ratio=0%2026.04 | 55.31 | 56.32 | 74.93 | 50.32 | 72.1 | 59.28 | 78.45 | 75.16 | — | — | 65.24 | |
| INTENTNoise=0%2026.04 | 55.31 | 56.32 | 74.93 | 50.32 | 72.1 | 59.28 | 78.45 | 75.16 | — | — | 65.24 | |
| SPRC + VQA4CIRbase_model=SPRC, enhancement=VQA4CIR2026.04 | 55.21 | 56.79 | 74.52 | 49.18 | 73.06 | 59.67 | 79.3 | 75.62 | — | — | 65.41 | |
| Air-KnowNoise Ratio=20%, Backbone=BLIP-22026.04 | 55.18 | 56.38 | 74.73 | 50.22 | 73.48 | 58.95 | 78.94 | 75.71 | — | — | 65.45 | |
| TMENoise=0%2026.04 | 55.16 | 56.43 | 74.44 | 49.73 | 71.69 | 59.31 | 78.94 | 75.02 | — | — | 65.09 | |
| TMENoise=0%2026.06 | 55.16 | 56.43 | 74.44 | 49.73 | 71.69 | 59.31 | 78.94 | 75.02 | — | — | 65.09 | |
| TMEVenue=CVPR’252026.02 | 55.15 | 56.43 | 74.44 | 49.73 | 71.69 | 59.31 | 78.94 | 75.02 | — | — | — | |
| TME2026.01 | 55.15 | 56.43 | 74.44 | 49.73 | 71.69 | 59.31 | 78.94 | 75.02 | — | — | 65.09 | |
| TMEBackbone=BLIP-2 ViT-G, Leverage additional data=false2026.04 | 55.15 | 56.43 | 74.44 | 49.73 | 71.69 | 59.31 | 78.94 | 75.02 | — | — | 65.09 | |
| TMENoise Ratio=0%2026.04 | 55.15 | 56.43 | 74.44 | 49.73 | 71.69 | 59.31 | 78.94 | 75.02 | — | — | 65.09 | |
| TMENoise=0%2026.04 | 55.15 | 56.43 | 74.44 | 49.73 | 71.69 | 59.31 | 78.94 | 75.02 | — | — | 65.09 | |
| ConeSepNoise=20%2026.04 | 54.93 | 56.28 | 74.45 | 50.23 | 72.19 | 58.29 | 78.38 | 75.01 | — | — | 64.97 | |
| SPRC2023.10 | 54.92 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| SPRC2025.04 | 54.92 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| SPRC2026.01 | 54.92 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| SPRCBackbone=BLIP-2 ViT-G, Leverage additional data=false2026.04 | 54.92 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| SPRCNoise Ratio=0%2026.04 | 54.92 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| SPRCNoise=0%2026.04 | 54.92 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| SPRCVenue=ICLR’242026.02 | 54.72 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | — | |
| SPRCNoise=0%2026.04 | 54.72 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| SPRCNoise=0%2026.06 | 54.72 | 55.64 | 73.89 | 49.18 | 72.43 | 59.35 | 78.58 | 74.97 | — | — | 64.85 | |
| RankVRNoise=0%2026.06 | 54.68 | 55.82 | 74.02 | 50.26 | 72.42 | 57.95 | 77.91 | 74.75 | — | — | 64.73 | |
| DQU-CIRBackbone=CLIP ViT-H, Leverage additional data=true2026.04 | 54.65 | 53.57 | 73.21 | 51.9 | 74.37 | 58.48 | 79.23 | 75.6 | — | — | 65.12 | |
| DQE-CIR2026.03 | 54.6 | 55.94 | 74.62 | 48.47 | 71.09 | 59.38 | 79.12 | 75.94 | — | — | — | |
| HABITNoise Ratio=20%2026.04 | 54.48 | 55.67 | 73.19 | 49.63 | 71.34 | 58.14 | 78.32 | 74.28 | — | — | 64.38 | |
| HABITNoise=20%2026.04 | 54.48 | 55.67 | 73.19 | 49.63 | 71.34 | 58.14 | 78.32 | 74.28 | — | — | 64.38 | |
| CCINVenue=CVPR’252026.02 | 54.41 | 55.93 | 74.14 | 49.38 | 72.58 | 57.93 | 77.56 | 74.76 | — | — | — | |
| CCIN2026.01 | 54.41 | 55.93 | 74.14 | 49.38 | 72.58 | 57.93 | 77.56 | 74.76 | — | — | 64.59 | |
| CCINBackbone=BLIP-2 ViT-G, Leverage additional data=false2026.04 | 54.41 | 55.93 | 74.14 | 49.38 | 72.58 | 57.93 | 77.56 | 74.76 | — | — | 64.59 | |
| SPRC2026.04 | 54.41 | 55.05 | 74.39 | 49.03 | 72.68 | 59.15 | 78.99 | 75.35 | — | — | 64.88 | |
| RankVRNoise=20%2026.06 | 54.36 | 55.97 | 73.22 | 49.81 | 71.48 | 57.29 | 77.94 | 74.21 | — | — | 64.29 | |
| INTENTNoise=20%2026.04 | 54.22 | 55.32 | 73.57 | 49.32 | 71.43 | 58.01 | 78.46 | 74.49 | — | — | 64.36 | |
| INTENTNoise Ratio=20%2026.04 | 54.22 | 55.32 | 73.57 | 49.32 | 71.43 | 58.01 | 78.46 | 74.49 | — | — | 64.36 | |
| INTENTNoise=20%2026.04 | 54.22 | 55.32 | 73.57 | 49.32 | 71.43 | 58.01 | 78.46 | 74.49 | — | — | 64.36 | |
| TMENoise=20%2026.04 | 54.03 | 55.84 | 73.16 | 49.03 | 70.35 | 57.22 | 78.23 | 73.91 | — | — | 63.97 | |
| TMENoise Ratio=20%2026.04 | 54.03 | 55.84 | 73.16 | 49.03 | 70.35 | 57.22 | 78.23 | 73.91 | — | — | 63.97 | |
| TMENoise=20%2026.04 | 54.03 | 55.84 | 73.16 | 49.03 | 70.35 | 57.22 | 78.23 | 73.91 | — | — | 63.97 | |
| TMENoise=20%2026.06 | 54.03 | 55.84 | 73.16 | 49.03 | 70.35 | 57.22 | 78.23 | 73.91 | — | — | 63.97 | |
| RCLNoise=0%2026.04 | 53.86 | 55.89 | 73.9 | 48.79 | 72.68 | 56.91 | 77.41 | 74.66 | — | — | 64.26 | |
| RCLNoise=0%2026.06 | 53.86 | 55.89 | 73.9 | 48.79 | 72.68 | 56.91 | 77.41 | 74.66 | — | — | 64.26 | |
| ConeSepNoise=50%2026.04 | 53.39 | 54.56 | 73.25 | 48.39 | 71.34 | 57.22 | 77.87 | 74.15 | — | — | 63.77 | |
| BaselineStage=Rbase2026.02 | 53.23 | 55 | 74.39 | 46.8 | 70.6 | 57.88 | 78.12 | 74.37 | — | — | — | |
| Air-KnowNoise Ratio=50%, Backbone=BLIP-22026.04 | 53.19 | 53.95 | 73.09 | 47.9 | 71.09 | 57.71 | 77.94 | 74.04 | — | — | 63.61 | |
| RDENoise=0%2026.04 | 53.04 | 54.37 | 73.55 | 47.84 | 71.89 | 56.91 | 77.21 | 74.22 | — | — | 63.63 | |
| RDENoise=0%2026.06 | 53.04 | 54.37 | 73.55 | 47.84 | 71.89 | 56.91 | 77.21 | 74.22 | — | — | 63.63 | |
| TEMAText Encoder=BLIP2026.04 | 53.02 | 52.9 | 73.55 | 49.66 | 71.98 | 56.49 | 77.07 | 74.2 | — | — | — | |
| QuReVenue=ICML’252026.02 | 52.6 | 53.53 | 72.87 | 46.8 | 69.81 | 57.47 | 77.77 | 73.48 | — | — | — | |
| QuRe2026.03 | 52.6 | 53.53 | 72.87 | 46.8 | 69.81 | 57.47 | 77.77 | 73.48 | — | — | — | |
| QuReNoise Ratio=0%2026.04 | 52.6 | 53.53 | 72.87 | 46.8 | 69.81 | 57.47 | 77.77 | 73.48 | — | — | 63.04 | |
| INTENTNoise=50%2026.04 | 52.52 | 52.78 | 72.48 | 47.99 | 71.24 | 56.79 | 76.23 | 73.32 | — | — | 62.92 | |
| HABITNoise Ratio=50%2026.04 | 52.52 | 53.72 | 72.55 | 47.33 | 69.71 | 56.51 | 77 | 73.09 | — | — | 62.8 | |
| INTENTNoise Ratio=50%2026.04 | 52.52 | 52.78 | 72.48 | 47.99 | 71.24 | 56.79 | 76.23 | 73.32 | — | — | 62.92 | |
| HABITNoise=50%2026.04 | 52.52 | 53.72 | 72.55 | 47.33 | 69.71 | 56.51 | 77 | 73.09 | — | — | 62.8 | |
| INTENTNoise=50%2026.04 | 52.52 | 52.78 | 72.48 | 47.99 | 71.24 | 56.79 | 76.23 | 73.32 | — | — | 62.92 | |
| RankVRNoise=50%2026.06 | 51.99 | 53.5 | 72.84 | 46.54 | 69.29 | 55.93 | 77.56 | 73.23 | — | — | 62.61 | |
| RCLNoise=20%2026.04 | 51.82 | 53.14 | 71.74 | 47.05 | 70.65 | 55.28 | 75.62 | 72.67 | — | — | 62.25 | |
| RCLNoise=20%2026.06 | 51.82 | 53.14 | 71.74 | 47.05 | 70.65 | 55.28 | 75.62 | 72.67 | — | — | 62.25 | |
| TMENoise=50%2026.04 | 51.47 | 53.09 | 71.88 | 46.26 | 68.27 | 55.07 | 76.59 | 72.25 | — | — | 61.86 | |
| TMENoise Ratio=50%2026.04 | 51.47 | 53.09 | 71.88 | 46.26 | 68.27 | 55.07 | 76.59 | 72.25 | — | — | 61.86 | |
| TMENoise=50%2026.04 | 51.47 | 53.09 | 71.88 | 46.26 | 68.27 | 55.07 | 76.59 | 72.25 | — | — | 61.86 | |
| TMENoise=50%2026.06 | 51.47 | 53.09 | 71.88 | 46.26 | 68.27 | 55.07 | 76.59 | 72.25 | — | — | 61.86 | |
| TG-CIRBackbone=CLIP2023.09 | 51.32 | 52.6 | 72.52 | 45.22 | 69.66 | 56.14 | 77.1 | 73.09 | — | — | — | |
| TG-CIR2023.10 | 51.32 | 52.6 | 72.52 | 45.22 | 69.66 | 56.14 | 77.1 | 73.09 | — | — | 62.21 | |
| TG-CIR2025.04 | 51.32 | 52.6 | 72.52 | 45.22 | 69.66 | 56.14 | 77.1 | 73.09 | — | — | 58.05 | |
| Re-ranking2023.10 | 51.17 | 50.15 | 71.25 | 48.14 | 71.43 | 55.23 | 76.8 | 73.13 | — | — | 62.15 | |
| Candidate Re-ranking for CIR (Ours R100)model_role=candidate re-ranking, top_k_filtered=1002023.05 | 51.17 | 50.15 | 71.25 | 48.14 | 71.34 | 55.23 | 76.8 | 73.13 | — | — | 62.15 | |
| Re-ranking2025.04 | 51.17 | 50.15 | 71.25 | 48.14 | 71.43 | 55.23 | 76.8 | 73.13 | — | — | 62.15 | |
| Re-ranking2026.01 | 51.17 | 50.15 | 71.25 | 48.14 | 71.43 | 55.23 | 76.8 | 73.13 | — | — | 62.15 | |
| Re-rankingBackbone=BLIP ViT-B, Leverage additional data=false2026.04 | 51.17 | 50.15 | 71.25 | 48.14 | 71.43 | 55.23 | 76.8 | 73.13 | — | — | 62.15 | |
| ReRanking2026.04 | 51.17 | 50.15 | 71.25 | 48.14 | 71.34 | 55.23 | 76.8 | 73.13 | — | — | 62.15 | |
| CandidateYear=2024, Text Encoder=BLIP2026.04 | 51.17 | 50.15 | 71.25 | 48.14 | 71.34 | 55.23 | 76.8 | 73.13 | — | — | — | |
| FashionERN2026.01 | 50.91 | 52.7 | 75.07 | 43.93 | 68.77 | 56.09 | 78.38 | 74.07 | — | — | 62.49 | |
| SPRC2026.03 | 50.86 | 51.37 | 72.77 | 45.71 | 70 | 55.48 | 77.46 | 73.41 | — | — | — | |
| CoVR-2Venue=TPAMI’242026.02 | 49.96 | 51.23 | 70.64 | 46.53 | 69.6 | 52.14 | 73.27 | 71.17 | — | — | — | |
| ConeSepNoise=80%2026.04 | 49.25 | 51.23 | 70.35 | 43.33 | 66.01 | 53.19 | 74.55 | 70.3 | — | — | 59.77 | |
| Air-KnowNoise Ratio=80%, Backbone=BLIP-22026.04 | 49.23 | 51.31 | 70.26 | 42.52 | 66.03 | 53.86 | 74.41 | 70.23 | — | — | 59.73 | |
| RDENoise=20%2026.04 | 49.16 | 50.74 | 69.09 | 44.62 | 68.91 | 52.12 | 73.38 | 70.64 | — | — | 59.81 | |
| RDENoise=20%2026.06 | 49.16 | 50.74 | 69.09 | 44.62 | 68.91 | 52.12 | 73.38 | 70.46 | — | — | 59.81 | |
| FoCoBackbone=ViT-G/142026.07 | 49.1 | 51.5 | 69.2 | 41.1 | 64.6 | 54.6 | 72.2 | 68.7 | — | — | — | |
| RCLNoise=50%2026.06 | 49.02 | 50.74 | 69.19 | 43.68 | 66.44 | 52.63 | 73.84 | 69.82 | — | — | 59.42 | |
| RCLNoise=50%2026.04 | 49.01 | 50.74 | 69.19 | 43.68 | 66.44 | 52.63 | 73.84 | 69.82 | — | — | 59.42 | |
| CASE2023.10 | 48.79 | 48.48 | 70.23 | 47.44 | 69.36 | 50.18 | 72.24 | 70.68 | — | — | 59.74 | |
| CASEbackbone=BLIP2023.03 | 48.79 | 48.48 | 70.23 | 47.77 | 69.36 | 50.18 | 72.24 | 70.68 | — | — | 59.74 |