Composed Image Retrieval on CIRR
94.8Recall@1ORACLE-Q
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T32026.05 | 94.8 | — | — | — | — | — | — | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T32026.05 | 91.3 | — | — | — | — | — | — | — | |
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T22026.05 | 90.2 | — | — | — | — | — | — | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T22026.05 | 85.6 | — | — | — | — | — | — | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T32026.05 | 80.1 | — | — | — | — | — | — | — | |
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T12026.05 | 79.5 | — | — | — | — | — | — | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T22026.05 | 74.8 | — | — | — | — | — | — | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T12026.05 | 74.2 | — | — | — | — | — | — | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T32026.05 | 71.2 | — | — | — | — | — | — | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T22026.05 | 67.9 | — | — | — | — | — | — | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T12026.05 | 66 | — | — | — | — | — | — | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T32026.05 | 63.4 | — | — | — | — | — | — | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T12026.05 | 63.1 | — | — | — | — | — | — | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T22026.05 | 62 | — | — | — | — | — | — | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T12026.05 | 60.2 | — | — | — | — | — | — | — | |
| RANDOM-QPolicy=RANDOM-Q, Turn budget (T)=T02026.05 | 58.4 | — | — | — | — | — | — | — | |
| FIXED-QPolicy=FIXED-Q, Turn budget (T)=T02026.05 | 58.4 | — | — | — | — | — | — | — | |
| MLLM-QPolicy=MLLM-Q, Turn budget (T)=T02026.05 | 58.4 | — | — | — | — | — | — | — | |
| Ours (EIG)Policy=Ours (EIG), Turn budget (T)=T02026.05 | 58.4 | — | — | — | — | — | — | — | |
| ORACLE-QPolicy=ORACLE-Q, Turn budget (T)=T02026.05 | 58.4 | — | — | — | — | — | — | — | |
| HINT2026.03 | 52.34 | 82.65 | 90.5 | 98.39 | 80.48 | 92.31 | 96.77 | 81.57 | |
| ReTrack2026.04 | 52.34 | 82.53 | 90.34 | 98.13 | 79.64 | 92.58 | 96.99 | — | |
| IMAGINEModel Category=CVR Models2026.06 | 52.25 | 82.39 | 90.34 | 98.12 | 80.14 | 92.43 | 96.92 | 84.66 | |
| QuRe2026.03 | 52.22 | 82.53 | 90.31 | 98.17 | 78.51 | 91.28 | 96.48 | 80.52 | |
| QuReModel Category=CIR Models2026.06 | 52.22 | 82.53 | 90.31 | 98.17 | 78.51 | 91.28 | 96.48 | 84.21 | |
| SPRC2026.03 | 51.96 | 82.12 | 89.74 | 97.69 | 80.65 | 92.31 | 96.6 | 81.39 | |
| SPRC2026.04 | 51.96 | 82.12 | 89.74 | 97.69 | 80.65 | 92.31 | 96.6 | — | |
| SPRCModel Category=CIR Models2026.06 | 51.96 | 82.12 | 89.74 | 97.69 | 80.65 | 92.31 | 96.6 | 84.44 | |
| OSCARType=Agentic, Training Free=true2026.02 | 51.18 | 79.5 | 87.45 | 96.56 | — | — | — | — | |
| CoVR Enrich2026.04 | 51.03 | — | 88.93 | 97.53 | 76.51 | — | 95.76 | — | |
| CoVR_EnrichModel Category=CVR Models2026.06 | 51.03 | — | 88.93 | 97.53 | 76.51 | — | 95.76 | — | |
| CoVR-22026.04 | 50.43 | 81.08 | 88.89 | 98.05 | 76.75 | 90.34 | 95.78 | — | |
| CoVR-2Model Category=CVR Models2026.06 | 50.43 | 81.08 | 88.89 | 98.05 | 76.75 | 90.34 | 95.78 | 83.95 | |
| CoVR2026.04 | 49.69 | 78.6 | 86.77 | 94.31 | 75.01 | 88.12 | 93.16 | — | |
| CoVRModel Category=CVR Models2026.06 | 49.69 | 78.6 | 86.77 | 94.31 | 75.01 | 88.12 | 93.16 | 80.81 | |
| COPE2026.03 | 49.18 | 80.65 | 89.86 | 98.05 | 72.34 | 88.65 | 95.3 | 76.49 | |
| CALA2026.03 | 49.11 | 81.21 | 89.59 | 98 | 76.27 | 91.04 | 96.46 | 78.74 | |
| PAIR2026.03 | 46.36 | 78.43 | 87.86 | 97.9 | 74.63 | 89.64 | 95.61 | 76.53 | |
| PAIRModel Category=CIR Models2026.06 | 46.36 | 78.43 | 87.86 | 97.9 | 74.63 | 89.64 | 95.61 | 81.49 | |
| ENCODER2026.04 | 46.1 | 77.98 | 87.16 | 97.64 | 76.92 | 90.41 | 95.95 | — | |
| ENCODERModel Category=CIR Models2026.06 | 46.1 | 77.98 | 87.16 | 97.64 | 76.92 | 90.41 | 95.95 | 81.74 | |
| MEDIAN2026.03 | 45.66 | 78.72 | 87.88 | 97.89 | 75.52 | 89.45 | 95.57 | 77.12 | |
| MEDIANModel Category=CIR Models2026.06 | 45.66 | 78.72 | 87.88 | 97.89 | 75.52 | 89.45 | 95.57 | 81.53 | |
| TG-CIR2026.03 | 45.25 | 78.29 | 87.16 | 97.3 | 72.84 | 89.25 | 95.13 | 75.57 | |
| TG-CIR2026.04 | 45.25 | 78.29 | 87.16 | 97.3 | 72.84 | 89.25 | 95.13 | — | |
| TG-CIRModel Category=CIR Models2026.06 | 45.25 | 78.29 | 87.16 | 97.3 | 72.84 | 89.25 | 95.13 | 80.75 | |
| SADN2026.04 | 44.27 | 78.1 | 87.71 | 97.89 | 72.34 | 88.7 | 95.23 | — | |
| SADNModel Category=CIR Models2026.06 | 44.27 | 78.1 | 87.71 | 97.89 | 72.34 | 88.7 | 95.23 | 80.61 | |
| SSN2026.03 | 43.91 | 77.25 | 86.48 | 97.45 | 71.76 | 88.63 | 95.54 | 74.51 | |
| SSN2026.04 | 43.91 | 77.25 | 86.48 | 97.45 | 71.76 | 88.63 | 95.54 | — | |
| MMF2026.03 | 43.8 | 77.37 | 86.94 | 97.71 | 71.39 | 88.16 | 95.15 | 74.38 | |
| LIMN2026.04 | 43.64 | 75.37 | 85.42 | 97.04 | 69.01 | 86.22 | 94.19 | — | |
| LIMNModel Category=CIR Models2026.06 | 43.64 | 75.37 | 85.42 | 97.04 | 69.01 | 86.22 | 94.19 | 78.8 | |
| FiREType=CIR Dedicated, Training Free=false2026.02 | 43.33 | 74.02 | 83.51 | 95.83 | — | — | — | — | |
| LIMN+2026.04 | 43.33 | 75.41 | 85.81 | 97.21 | 69.28 | 86.43 | 94.26 | — | |
| LIMN+Model Category=CIR Models2026.06 | 43.33 | 75.41 | 85.81 | 97.21 | 69.28 | 86.43 | 94.26 | 78.79 | |
| X^RType=Agentic, Training Free=true2026.02 | 43.13 | 73.59 | 83.09 | 94.05 | — | — | — | — | |
| Baseline (ViT-G)Pretraining Data=OACIRR, Pretraining Scale=127 K, Zero-shot=true2026.04 | 42.96 | 73.62 | — | — | 72.95 | — | — | 73.28 | |
| Baseline (ViT-L)Pretraining Data=OACIRR, Pretraining Scale=127 K, Zero-shot=true2026.04 | 41.59 | 72.36 | — | — | 72.06 | — | — | 72.21 | |
| CoAlign (ViT-G)Pretraining Data=CIRHS, Pretraining Scale=535 K, Zero-shot=true2026.04 | 41.08 | 71.11 | — | — | 70.8 | — | — | 70.96 | |
| CoVRBackbone=BLIP ViT-L/142024.06 | 39.3 | 68.2 | 78.9 | — | — | — | — | — | |
| ParacosmBackbone=OpenCLIP ViT-G/142026.01 | 39.3 | — | — | — | — | — | — | — | |
| LDRE + IP-CIRBackbone=ViT-G/142024.11 | 39.25 | 70.07 | 80 | 94.89 | 69.95 | 86.87 | 94.22 | — | |
| CoVR-BLIPPretraining Data=WebVid-CoVR, Pretraining Scale=1,644 K, Zero-shot=true2026.04 | 38.48 | 66.7 | — | — | 69.28 | — | — | 67.99 | |
| MRA-CIRType=Agentic, Training Free=false2026.02 | 37.98 | 67.45 | 78.07 | 93.98 | — | — | — | — | |
| OSrCIRBackbone=OpenCLIP ViT-G/14, reproduced=true2026.01 | 37.59 | — | — | — | — | — | — | — | |
| LinCIR + RTDBackbone=CLIP ViT-G/14, Integration with RTD=true2024.06 | 36.3 | 67.5 | 78.3 | — | — | — | — | — | |
| LDREBackbone=CLIP ViT-G/142024.06 | 36.2 | 66.4 | 77.3 | — | — | — | — | — | |
| LDREBackbone=ViT-G/142024.11 | 36.15 | 66.39 | 77.25 | 93.95 | 68.82 | 85.66 | 93.76 | — | |
| LinCIR + RTDBackbone=FSC-CLIP ViT-L/14, Integration with RTD=true2024.06 | 35.9 | 67.5 | 78.1 | — | — | — | — | — | |
| GRAPEBackbone=ViT-B/16, Method=+GRAPE2025.09 | 35.7 | — | 77.8 | — | — | — | — | — | |
| CASEBackbone=BLIP ViT-L/142024.06 | 35.4 | 65.8 | 78.5 | — | — | — | — | — | |
| CASEPretraining Data=LaSCo + CoCo, Pretraining Scale=389 K, Zero-shot=true2026.04 | 35.4 | 65.78 | — | — | 64.29 | — | — | 65.04 | |
| LinCIR + IP-CIRBackbone=ViT-G/142024.11 | 35.37 | 64.7 | 76.15 | 93.71 | 62.58 | 81.74 | 91.35 | — | |
| LinCIRBackbone=CLIP ViT-G/14, Integration with RTD=false2024.06 | 34.9 | 64.5 | 76.1 | — | — | — | — | — | |
| SEARLEBackbone=OpenCLIP ViT-G/142026.01 | 34.8 | — | — | — | — | — | — | — | |
| LinCIRBackbone=ViT-G/142024.11 | 34.75 | 64.12 | 75.93 | 93.42 | 62.36 | 81.78 | 91.28 | — | |
| CompodiffBackbone=CLIP ViT-G/142024.06 | 34.7 | 64.3 | 75.1 | — | — | — | — | — | |
| CIREVLBackbone=ViT-G/142024.11 | 34.65 | 64.29 | 75.06 | 91.66 | 67.95 | 84.87 | 93.21 | — | |
| CIReVLBackbone=OpenCLIP ViT-G/142026.01 | 34.65 | — | — | — | — | — | — | — | |
| GRAPEBackbone=ViT-B/32, Method=+GRAPE2025.09 | 34.3 | — | 77.8 | — | — | — | — | — | |
| LinCIRBackbone=FSC-CLIP ViT-L/14, Integration with RTD=false2024.06 | 33.6 | 63.1 | 74.5 | — | — | — | — | — | |
| MagicLens-LBackbone=CoCa-L, # Total Params=613M2024.03 | 33.3 | — | — | — | — | — | — | — | |
| GRAPEBackbone=ViT-L/14, Method=+GRAPE2025.09 | 33.2 | — | 76.3 | — | — | — | — | — | |
| AutoCIRType=Agentic, Training Free=true2026.02 | 31.81 | 61.95 | 73.86 | 92.07 | — | — | — | — | |
| MagicLens-BBackbone=CoCa-B, # Total Params=267M2024.03 | 31.6 | — | — | — | — | — | — | — | |
| CLIP + LLMBackbone=ViT-B/16, Method=+LLM2025.09 | 30.7 | — | 73.3 | — | — | — | — | — | |
| Pic2WordBackbone=OpenCLIP ViT-G/142026.01 | 30.41 | — | — | — | — | — | — | — | |
| MagicLens-LBackbone=CLIP-L, # Total Params=465M2024.03 | 30.1 | — | — | — | — | — | — | — | |
| MagicLensBackbone=CLIP ViT-L/142024.06 | 30.1 | 61.7 | 74.4 | — | — | — | — | — | |
| MagicLensBackbone=ViT-L/14, Training Costs=128 TPU, 6h2026.07 | 30.1 | 61.7 | 74.4 | — | — | — | — | — | |
| LDRE + IP-CIRBackbone=ViT-L/142024.11 | 29.76 | 58.82 | 71.21 | 90.41 | 62.48 | 81.64 | 90.89 | — | |
| CLIP + LLMBackbone=ViT-L/14, Method=+LLM2025.09 | 29.7 | — | 70.1 | — | — | — | — | — | |
| CLARASetting=Zero-shot, Turns=m=02026.06 | 29.7 | 61.6 | 76.1 | — | — | — | — | — | |
| CLIP + LLMBackbone=ViT-B/32, Method=+LLM2025.09 | 29.5 | — | 70.5 | — | — | — | — | — | |
| OSrCIRSetting=Zero-shot, Turns=m=02026.06 | 29.4 | 62 | 75.8 | — | — | — | — | — | |
| Context-I2W + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 29.2 | 58.4 | 70.5 | — | — | — | — | — | |
| FTI4CIR + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 29.1 | 58.8 | 70.4 | — | — | — | — | — | |
| QQMM-embed-v2Type=Multimodal Embedding, Training Free=true2026.02 | 28.98 | 73.66 | 82.98 | 96.72 | — | — | — | — | |
| Pic2Word + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 27.9 | 56.2 | 68.5 | — | — | — | — | — |