Composed Image Retrieval on Fashion-IQ
82.01Average Recall@50MCoT-MVS
Evaluation Results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MCoT-MVSMethod category=VLP Model-Based Methods2026.03 | 82.01 | 58.45 | 78.92 | 63.24 | 81.15 | 68.02 | 85.97 | 63.24 | — | — | — | — | — | — | — | — | — | |
| DQU-CIRMethod category=VLP Model-Based Methods, Venue=SIGIR'242026.03 | 81.56 | 57.63 | 78.56 | 62.14 | 80.38 | 66.15 | 85.73 | 61.97 | — | — | — | — | — | — | — | — | — | |
| HINT2026.03 | 77.93 | 51.8 | 73.13 | 60.27 | 78.48 | 63.37 | 82.19 | 58.48 | — | — | — | — | — | — | — | — | — | |
| ENCODERMethod category=VLP Model-Based Methods, Venue=AAAI'252026.03 | 77.59 | 51.51 | 76.95 | 54.86 | 74.93 | 62.01 | 80.88 | 56.13 | — | — | — | — | — | — | — | — | — | |
| MAPNet2026.06 | 76.2 | 51.17 | 74.12 | 56.37 | 75.17 | 59.56 | 79.3 | 55.7 | — | — | — | — | — | — | — | — | — | |
| CIRLVLMMethod category=VLP Model-Based Methods, Venue=AAAI'252026.03 | 76.14 | 50.42 | 73.57 | 58.59 | 75.86 | 59.61 | 78.99 | 56.21 | — | — | — | — | — | — | — | — | — | |
| CLIP-4CIRCaptioning=LLaVA-enhanced2026.06 | 75.67 | 49.62 | 71.83 | 57.02 | 75.66 | 59.31 | 79.5 | 55.32 | — | — | — | — | — | — | — | — | — | |
| SPRC2026.06 | 75.35 | 47.8 | 72.7 | 55.84 | 74.37 | 58.89 | 78.99 | 54.17 | — | — | — | — | — | — | — | — | — | |
| HABITNoise=0%2026.04 | 75.2 | 49.99 | 72.38 | 56.62 | 74.68 | 59.51 | 78.53 | 55.38 | — | — | 65.29 | — | — | — | — | — | — | |
| TMENoise=0%2026.04 | 75.02 | 49.73 | 71.69 | 56.43 | 74.44 | 59.31 | 78.94 | 55.15 | — | — | 65.09 | — | — | — | — | — | — | |
| SPRCMethod category=VLP Model-Based Methods, Venue=ICLR'242026.03 | 74.97 | 49.18 | 72.43 | 55.64 | 73.89 | 59.35 | 78.58 | 54.72 | — | — | — | — | — | — | — | — | — | |
| SPRC2026.03 | 74.97 | 49.18 | 72.43 | 55.64 | 73.89 | 59.35 | 78.58 | 54.72 | — | — | — | — | — | — | — | — | — | |
| SPRCNoise=0%2026.04 | 74.97 | 49.18 | 72.43 | 55.64 | 73.89 | 59.35 | 78.58 | 54.92 | — | — | 64.85 | — | — | — | — | — | — | |
| CCINMethod category=VLP Model-Based Methods, Venue=CVPR'252026.03 | 74.76 | 49.38 | 72.58 | 55.93 | 74.14 | 57.93 | 77.56 | 54.41 | — | — | — | — | — | — | — | — | — | |
| RCLNoise=0%2026.04 | 74.66 | 48.79 | 72.68 | 55.89 | 73.9 | 56.91 | 77.41 | 53.86 | — | — | 64.26 | — | — | — | — | — | — | |
| PAIR2026.03 | 74.51 | 46.78 | 70.93 | 52.6 | 73.8 | 58.91 | 78.81 | 52.76 | — | — | — | — | — | — | — | — | — | |
| MEDIAN2026.03 | 74.3 | 46.9 | 70.3 | 52.65 | 73.96 | 57.62 | 78.63 | 52.39 | — | — | — | — | — | — | — | — | — | |
| HABITNoise=20%2026.04 | 74.28 | 49.63 | 71.34 | 55.67 | 73.19 | 58.14 | 78.32 | 54.48 | — | — | 64.38 | — | — | — | — | — | — | |
| RDENoise=0%2026.04 | 74.22 | 47.84 | 71.89 | 54.37 | 73.55 | 56.91 | 77.21 | 53.04 | — | — | 63.63 | — | — | — | — | — | — | |
| TMENoise=20%2026.04 | 73.91 | 49.03 | 70.35 | 55.84 | 73.16 | 57.22 | 78.23 | 54.03 | — | — | 63.97 | — | — | — | — | — | — | |
| QuRe2026.03 | 73.48 | 46.8 | 69.81 | 53.53 | 72.87 | 57.47 | 77.77 | 52.6 | — | — | — | — | — | — | — | — | — | |
| TGCIRMethod category=VLP Model-Based Methods, Venue=ACM MM'232026.03 | 73.09 | 45.22 | 69.66 | 52.6 | 72.52 | 56.14 | 77.1 | 51.32 | — | — | — | — | — | — | — | — | — | |
| TG-CIR2026.03 | 73.09 | 45.22 | 69.66 | 52.6 | 72.52 | 56.14 | 77.1 | 51.32 | — | — | — | — | — | — | — | — | — | |
| HABITNoise=50%2026.04 | 73.09 | 47.33 | 69.71 | 53.72 | 72.55 | 56.51 | 77 | 52.52 | — | — | 62.8 | — | — | — | — | — | — | |
| FashionERNMethod category=VLP Model-Based Methods, Venue=AAAI'242026.03 | 72.95 | 50.32 | 71.29 | 50.15 | 70.36 | 56.4 | 77.21 | 52.29 | — | — | — | — | — | — | — | — | — | |
| RCLNoise=20%2026.04 | 72.67 | 47.05 | 70.65 | 53.14 | 71.74 | 55.28 | 75.62 | 51.82 | — | — | 62.25 | — | — | — | — | — | — | |
| TMENoise=50%2026.04 | 72.25 | 46.26 | 68.27 | 53.09 | 71.88 | 55.07 | 76.59 | 51.47 | — | — | 61.86 | — | — | — | — | — | — | |
| RDENoise=20%2026.04 | 70.64 | 44.62 | 68.91 | 50.74 | 69.09 | 52.12 | 73.38 | 49.16 | — | — | 59.81 | — | — | — | — | — | — | |
| CASEMethod category=VLP Model-Based Methods, Venue=AAAI'242026.03 | 70.61 | 47.44 | 69.36 | 48.48 | 70.23 | 50.18 | 72.24 | 48.7 | — | — | — | — | — | — | — | — | — | |
| CoVRBLIPMethod category=VLP Model-Based Methods, Venue=AAAI'242026.03 | 70.25 | 44.55 | 69.03 | 48.43 | 67.42 | 52.6 | 74.31 | 48.53 | — | — | — | — | — | — | — | — | — | |
| RCLNoise=50%2026.04 | 69.82 | 43.68 | 66.44 | 50.74 | 69.19 | 52.63 | 73.84 | 49.01 | — | — | 59.42 | — | — | — | — | — | — | |
| HABITNoise=80%2026.04 | 69.53 | 42.04 | 65.2 | 50.12 | 69.77 | 52.92 | 73.61 | 48.36 | — | — | 58.94 | — | — | — | — | — | — | |
| CaLaMethod category=VLP Model-Based Methods, Venue=SIGIR'242026.03 | 69.22 | 42.38 | 66.08 | 46.76 | 68.16 | 50.93 | 73.42 | 46.69 | — | — | — | — | — | — | — | — | — | |
| CALA2026.03 | 69.22 | 42.38 | 66.08 | 46.76 | 68.16 | 50.93 | 73.42 | 46.69 | — | — | — | — | — | — | — | — | — | |
| CALANoise=0%2026.04 | 69.22 | 42.38 | 66.08 | 46.76 | 68.16 | 50.93 | 73.42 | 46.69 | — | — | 57.96 | — | — | — | — | — | — | |
| COPE2026.03 | 68.6 | 39.85 | 66.98 | 45.03 | 66.81 | 48.61 | 72.01 | 44.5 | — | — | — | — | — | — | — | — | — | |
| TMENoise=80%2026.04 | 68.6 | 41.45 | 64.35 | 47.3 | 68.2 | 51.25 | 73.23 | 46.67 | — | — | 57.63 | — | — | — | — | — | — | |
| MMF2026.03 | 67.52 | 40.45 | 65.29 | 44.85 | 66.29 | 49.26 | 70.98 | 44.85 | — | — | — | — | — | — | — | — | — | |
| SADNMethod category=VLP Model-Based Methods, Venue=ACM MM'242026.03 | 67.36 | 40.01 | 65.1 | 43.67 | 66.05 | 48.04 | 70.93 | 43.91 | — | — | — | — | — | — | — | — | — | |
| LinCIR + RTDBackbone=CLIP ViT-G/14, Integration with RTD=true2024.06 | 67.3 | — | — | — | — | — | — | 46.2 | — | — | — | — | — | — | — | — | — | |
| RDENoise=50%2026.04 | 67.24 | 41.3 | 64.75 | 47.06 | 66.34 | 50.13 | 70.63 | 46.16 | — | — | 56.7 | — | — | — | — | — | — | |
| SPRCNoise=20%2026.04 | 66.36 | 39.81 | 62.22 | 48.58 | 66.29 | 50.48 | 70.58 | 46.29 | — | — | 56.33 | — | — | — | — | — | — | |
| SEIZEArch=ViT-G/14, Train=false, Pre-trained Weights Source=OpenCLIP2026.05 | 65.85 | 39.61 | 61.02 | 43.6 | 65.42 | 45.94 | 71.12 | 43.05 | — | — | — | — | — | — | — | — | — | |
| LinCIRArch=ViT-G/14, Train=true, Pre-trained Weights Source=OpenCLIP2026.05 | 65.69 | 38.08 | 60.88 | 46.76 | 65.11 | 50.48 | 71.09 | 45.11 | — | — | — | — | — | — | — | — | — | |
| LinCIRBackbone=CLIP ViT-G/14, Integration with RTD=false2024.06 | 65.5 | — | — | — | — | — | — | 44.5 | — | — | — | — | — | — | — | — | — | |
| IUDC2026.03 | 64.88 | 35.22 | 61.9 | 41.86 | 63.52 | 42.19 | 69.23 | 39.76 | — | — | — | — | — | — | — | — | — | |
| RCLNoise=80%2026.04 | 64.43 | 38.82 | 60.54 | 45.44 | 64.38 | 47.42 | 68.38 | 43.89 | — | — | 54.16 | — | — | — | — | — | — | |
| SSNMethod category=VLP Model-Based Methods, Venue=AAAI'242026.03 | 63.89 | 34.36 | 60.78 | 38.13 | 61.83 | 44.26 | 69.05 | 38.92 | — | — | — | — | — | — | — | — | — | |
| SSN2026.03 | 63.89 | 34.36 | 60.78 | 38.13 | 61.83 | 44.26 | 69.05 | 38.92 | — | — | — | — | — | — | — | — | — | |
| SSNNoise=0%2026.04 | 63.89 | 34.36 | 60.78 | 38.13 | 61.83 | 44.26 | 69.05 | 38.92 | — | — | 51.4 | — | — | — | — | — | — | |
| RDENoise=80%2026.04 | 62.75 | 37.63 | 59.64 | 43.62 | 62.12 | 46.1 | 66.5 | 42.45 | — | — | 52.6 | — | — | — | — | — | — | |
| Baseline (ViT-G)Pretraining Data=OACIRR, Pretraining Scale=127 K, Zero-shot=true2026.04 | 61.81 | — | — | — | — | — | — | 39.8 | — | — | — | — | — | — | — | — | — | |
| CLIP4CIRMethod category=VLP Model-Based Methods, Venue=TOMM'232026.03 | 61.74 | 33.81 | 59.4 | 39.99 | 60.45 | 41.41 | 65.37 | 38.4 | — | — | — | — | — | — | — | — | — | |
| SPRCNoise=50%2026.04 | 61.19 | 35.94 | 57.16 | 42.25 | 61.63 | 44.98 | 64.76 | 41.06 | — | — | 51.12 | — | — | — | — | — | — | |
| CoAlign (ViT-G)Pretraining Data=CIRHS, Pretraining Scale=535 K, Zero-shot=true2026.04 | 60.08 | — | — | — | — | — | — | 39.22 | — | — | — | — | — | — | — | — | — | |
| Baseline (ViT-L)Pretraining Data=OACIRR, Pretraining Scale=127 K, Zero-shot=true2026.04 | 59.57 | — | — | — | — | — | — | 37.82 | — | — | — | — | — | — | — | — | — | |
| STiTch(Ours)Arch=ViT-G/14, Train=false, Pre-trained Weights Source=OpenCLIP2026.05 | 59.43 | 35.04 | 56.74 | 39.48 | 56.59 | 42.86 | 64.95 | 39.12 | — | — | — | — | — | — | — | — | — | |
| CompoDiff2026.06 | 58.57 | 40.65 | 57.14 | 36.87 | 57.39 | 43.93 | 61.17 | 40.48 | — | — | — | — | — | — | — | — | — | |
| MT-CIRBackbone=CLIP ViT-L/142024.06 | 57.4 | — | — | — | — | — | — | 35.4 | — | — | — | — | — | — | — | — | — | |
| PLIBackbone=CLIP-L, # Params=428M, # Triplets=695K, Zero-shot=true2025.11 | 57.4 | — | — | — | — | — | — | 35.4 | — | — | — | — | — | — | — | — | — | |
| OSrCIR†Arch=ViT-G/14, Train=false, Pre-trained Weights Source=OpenCLIP (rerun)2026.05 | 57.15 | 30.59 | 53.5 | 37.39 | 56.92 | 39.72 | 61.04 | 35.79 | — | — | — | — | — | — | — | — | — | |
| OSrCIRArch=ViT-G/14, Train=false, Pre-trained Weights Source=OpenCLIP2026.05 | 57.11 | 33.02 | 54.78 | 38.65 | 54.71 | 41.04 | 61.83 | 37.57 | — | — | — | — | — | — | — | — | — | |
| PVLF#Triplets=32K, Zero-shot=true, Backbone=BLIP-base2025.11 | 56.9 | — | — | — | — | — | — | 35.7 | — | — | — | — | — | — | — | — | — | |
| OSrCIR*Arch=ViT-G/14, Train=false, Pre-trained Weights Source=OpenAI (rerun)2026.05 | 56.67 | 30.69 | 53.25 | 36.56 | 55.45 | 40.13 | 61.3 | 35.79 | — | — | — | — | — | — | — | — | — | |
| CoLLMBackbone=BLIP-B, # Triplets=3.4M, Zero-shot=true2025.11 | 56 | — | — | — | — | — | — | 34.6 | — | — | — | — | — | — | — | — | — | |
| CoLLM#Triplets=3.4M, Zero-shot=true, Backbone=BLIP-base2025.11 | 56 | — | — | — | — | — | — | 34.6 | — | — | — | — | — | — | — | — | — | |
| SEARLEArch=ViT-G/14, Train=true, Pre-trained Weights Source=OpenCLIP2026.05 | 55.71 | 28.16 | 50.32 | 36.46 | 55.35 | 39.83 | 61.45 | 34.81 | — | — | — | — | — | — | — | — | — | |
| LDREBackbone=CLIP ViT-G/142024.06 | 55.4 | — | — | — | — | — | — | 32.5 | — | — | — | — | — | — | — | — | — | |
| CALANoise=20%2026.04 | 55.28 | 29.05 | 51.36 | 35.28 | 56.23 | 36.05 | 58.24 | 33.46 | — | — | 44.37 | — | — | — | — | — | — | |
| TransAggBackbone=BLIP-B, # Params=235M, # Triplets=32K, Zero-shot=true2025.11 | 55.1 | — | — | — | — | — | — | 34.4 | — | — | — | — | — | — | — | — | — | |
| TransAgg#Triplets=32K, Zero-shot=true, Backbone=BLIP-base2025.11 | 55.1 | — | — | — | — | — | — | 34.4 | — | — | — | — | — | — | — | — | — | |
| OSrCIR†Arch=ViT-B/32, Train=false, Pre-trained Weights Source=OpenCLIP (rerun)2026.05 | 54.07 | 29.75 | 51.91 | 32.83 | 52.06 | 36.31 | 58.24 | 32.96 | — | — | — | — | — | — | — | — | — | |
| SPRCNoise=80%2026.04 | 54.03 | 28.41 | 50.77 | 36.21 | 54.37 | 35.9 | 59.06 | 33.51 | — | — | 43.77 | — | — | — | — | — | — | |
| OSrCIRArch=ViT-B/32, Train=false, Pre-trained Weights Source=OpenAI2026.05 | 53.4 | 29.35 | 50.37 | 31.16 | 51.13 | 36.51 | 58.71 | 32.34 | — | — | — | — | — | — | — | — | — | |
| MagicLensBackbone=CLIP ViT-L/142024.06 | 52.5 | — | — | — | — | — | — | 30.7 | — | — | — | — | — | — | — | — | — | |
| MagicLensBackbone=CLIP-L, # Params=465M, # Triplets=36.5M, Zero-shot=true2025.11 | 52.5 | — | — | — | — | — | — | 30.7 | — | — | — | — | — | — | — | — | — | |
| CIReVLBackbone=CLIP ViT-G/142024.06 | 52.4 | — | — | — | — | — | — | 32.2 | — | — | — | — | — | — | — | — | — | |
| CIReVLArch=ViT-G/14, Train=false, Pre-trained Weights Source=OpenCLIP2026.05 | 52.36 | 27.07 | 49.53 | 29.85 | 51.07 | 35.8 | 56.14 | 32.19 | — | — | — | — | — | — | — | — | — | |
| LinCIR + RTDBackbone=FSC-CLIP ViT-L/14, Integration with RTD=true2024.06 | 52.3 | — | — | — | — | — | — | 31.2 | — | — | — | — | — | — | — | — | — | |
| Pic2WordArch=ViT-G/14, Train=true, Pre-trained Weights Source=OpenCLIP2026.05 | 51.89 | 25.43 | 47.65 | 33.17 | 50.39 | 35.24 | 57.62 | 31.28 | — | — | — | — | — | — | — | — | — | |
| CompoDiffPretraining Data=ST18M + LAION-2B, Pretraining Scale=18,000 K, Zero-shot=true2026.04 | 51.71 | — | — | — | — | — | — | 39.02 | — | — | — | — | — | — | — | — | — | |
| FTI4CIR + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 51.7 | — | — | — | — | — | — | 30.6 | — | — | — | — | — | — | — | — | — | |
| CompodiffBackbone=CLIP ViT-G/142024.06 | 51.7 | — | — | — | — | — | — | 39 | — | — | — | — | — | — | — | — | — | |
| TransAgg + UNIONBackbone=BLIP-B, # Params=235M, # Triplets=5K, Zero-shot=true2025.11 | 51.5 | — | — | — | — | — | — | 31.9 | — | — | — | — | — | — | — | — | — | |
| LinCIR + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 51.1 | — | — | — | — | — | — | 30.2 | — | — | — | — | — | — | — | — | — | |
| FTI4CIRBackbone=CLIP ViT-L/14, Integration with RTD=false2024.06 | 50.9 | — | — | — | — | — | — | 29.4 | — | — | — | — | — | — | — | — | — | |
| SEARLE + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 50.7 | — | — | — | — | — | — | 29.3 | — | — | — | — | — | — | — | — | — | |
| LDREBackbone=CLIP ViT-L/142024.06 | 50.5 | — | — | — | — | — | — | 28.5 | — | — | — | — | — | — | — | — | — | |
| ARTEMISMethod category=Traditional Model-Based Methods, Venue=ICLR'222026.03 | 50.29 | 27.16 | 52.4 | 21.78 | 43.64 | 29.2 | 54.83 | 26.05 | — | — | — | — | — | — | — | — | — | |
| MLLM-I2WBackbone=CLIP-L, # Triplets=3M, Zero-shot=true2025.11 | 50.1 | — | — | — | — | — | — | 30.3 | — | — | — | — | — | — | — | — | — | |
| CIReVL†Arch=ViT-G/14, Train=false, Pre-trained Weights Source=OpenCLIP (rerun)2026.05 | 50.1 | 25.09 | 45.12 | 32.63 | 50.05 | 34.42 | 55.12 | 30.71 | — | — | — | — | — | — | — | — | — | |
| SSNNoise=20%2026.04 | 49.81 | 22.61 | 45.56 | 27.87 | 48.58 | 31.82 | 55.28 | 27.43 | — | — | 38.62 | — | — | — | — | — | — | |
| Context-I2W + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 49.5 | — | — | — | — | — | — | 28.1 | — | — | — | — | — | — | — | — | — | |
| LinCIRBackbone=FSC-CLIP ViT-L/14, Integration with RTD=false2024.06 | 49.5 | — | — | — | — | — | — | 28.3 | — | — | — | — | — | — | — | — | — | |
| i-SEARLEBackbone=CLIP-L, # Params=442M, # Triplets=205K, Zero-shot=true2025.11 | 49.5 | — | — | — | — | — | — | 29.2 | — | — | — | — | — | — | — | — | — | |
| CIReVLArch=ViT-B/32, Train=false, Pre-trained Weights Source=OpenAI2026.05 | 49.35 | 25.29 | 46.36 | 28.36 | 47.84 | 31.21 | 53.85 | 28.28 | — | — | — | — | — | — | — | — | — | |
| SEIZEArch=ViT-B/32, Train=false, Pre-trained Weights Source=OpenAI2026.05 | 49.35 | 25.37 | 46.84 | 29.38 | 47.97 | 32.07 | 54.78 | 28.94 | — | — | — | — | — | — | — | — | — | |
| Context-I2WBackbone=CLIP ViT-L/14, Integration with RTD=false2024.06 | 49.1 | — | — | — | — | — | — | 27.9 | — | — | — | — | — | — | — | — | — | |
| Pic2Word + RTDBackbone=CLIP ViT-L/14, Integration with RTD=true2024.06 | 48.9 | — | — | — | — | — | — | 27.6 | — | — | — | — | — | — | — | — | — | |
| VaGFeM_Triplet#Triplets=10K, Zero-shot=true, Backbone=BLIP-base2025.11 | 48.8 | — | — | — | — | — | — | 28.7 | — | — | — | — | — | — | — | — | — |