Composed Image Retrieval on PinPoint (test)
29mAP@10MMRet-MLLM-S1 + Reranking
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MMRet-MLLM-S1 + RerankingMethod Family=CIR Family, Backbone=MLLM-S1, Reranking=true2026.03 | 29 | 29.5 | 2.01 | 5.6 | 19.1 | |
| GPT-5-Text – Premerge + RerankingMethod Family=Text Generation, Fusion Strategy=Premerge, Reranking=true2026.03 | 27.2 | 28.1 | 3.46 | 6.2 | 19.3 | |
| GPT-5-Text – PremergeMethod Family=Text Generation, Fusion Strategy=Premerge2026.03 | 26.6 | 28.5 | 6.93 | 9 | 17.4 | |
| GPT-5-Text – PostmergeMethod Family=Text Generation, Fusion Strategy=Postmerge2026.03 | 23.6 | 25.5 | 7.94 | 9.4 | 18 | |
| MMRet-CLIP-L + RerankingMethod Family=CIR Family, Backbone=CLIP-L, Reranking=true2026.03 | 23.6 | 24.4 | 3.45 | 7.4 | 20.8 | |
| MagicLens-CLIP-L + RerankingMethod Family=CIR Family, Backbone=CLIP-L, Reranking=true2026.03 | 23.1 | 23.8 | 3.14 | 7.8 | 20.1 | |
| MMRet-MLLM-S1Method Family=CIR Family, Backbone=MLLM-S12026.03 | 22.4 | 23.8 | 6.38 | 9.1 | 16.2 | |
| MMRet-MLLM-S2Method Family=CIR Family, Backbone=MLLM-S22026.03 | 20.3 | 23.9 | 18.03 | 15.1 | 14.9 | |
| MMRet-CLIP-LMethod Family=CIR Family, Backbone=CLIP-L2026.03 | 17.8 | 19.7 | 10.89 | 12 | 18.8 | |
| MagicLens-CLIP-LMethod Family=CIR Family, Backbone=CLIP-L2026.03 | 15.5 | 17.8 | 14.41 | 15.1 | 18.2 | |
| MMRet-CLIP-BMethod Family=CIR Family, Backbone=CLIP-B2026.03 | 15.3 | 17.3 | 12.72 | 12.3 | 17.6 | |
| MagicLens-CLIP-BMethod Family=CIR Family, Backbone=CLIP-B2026.03 | 11.1 | 12.8 | 15.78 | 15.9 | 15.2 | |
| LinCIRMethod Family=CIR Family2026.03 | 11 | 13.5 | 23.47 | 14.1 | 15.2 | |
| Meta CLIP 2 – Combined + RerankingMethod Family=CLIP Family, Fusion Strategy=Combined, Reranking=true2026.03 | 8.7 | 8.9 | 2.71 | 3.9 | 14.9 | |
| Pic2WordMethod Family=CIR Family2026.03 | 4.6 | 6.8 | 45.63 | 15.3 | 10.6 | |
| Meta CLIP 2 – CombinedMethod Family=CLIP Family, Fusion Strategy=Combined2026.03 | 4.4 | 6.2 | 39.87 | 7.2 | 11.4 | |
| Meta CLIP 2 – SLERPMethod Family=CLIP Family, Fusion Strategy=SLERP2026.03 | 4.3 | 6.5 | 48.78 | 8.4 | 11.5 | |
| Meta CLIP 2 – Text OnlyMethod Family=CLIP Family, Input Modality=Text Only2026.03 | 3.1 | 3.3 | 4.95 | 1.8 | 8.7 | |
| Meta CLIP 2 – Image OnlyMethod Family=CLIP Family, Input Modality=Image Only2026.03 | 1.7 | 3 | 77.09 | 16.4 | 3.7 |