Composed Image Retrieval on GeneCIS (test)
17.9Recall@1OSrCIR w/ GPT-4o
Evaluation Results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OSrCIR w/ GPT-4oBackbone=CLIP ViT-L/14, venue&year=CVPR’252026.01 | 17.9 | 29 | 38.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoTMR w/ Qwen2.5-VLBackbone=OpenCLIP ViT-B/32, venue&year=ICCV’252026.01 | 17.9 | 29.6 | 39.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoTMR w/ Qwen2.5-VLBackbone=OpenCLIP ViT-L/14, venue&year=ICCV’252026.01 | 17.6 | 29.6 | 39.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ParacosmBackbone=OpenCLIP ViT-L/14, venue&year=ours2026.01 | 17.6 | 29 | 38.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ GPT-4oBackbone=CLIP ViT-B/32, venue&year=CVPR’252026.01 | 17.4 | 29.1 | 39 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ParacosmBackbone=CLIP ViT-L/14, venue&year=ours2026.01 | 17.4 | 28.5 | 37.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ParacosmBackbone=OpenCLIP ViT-B/32, venue&year=ours2026.01 | 17.1 | 28.6 | 38.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ParacosmBackbone=CLIP ViT-B/32, venue&year=ours2026.01 | 16.1 | 26.3 | 36.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CIReVLBackbone=CLIP ViT-B/32, venue&year=ICLR’242026.01 | 15.9 | 26.8 | 36.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CIReVLBackbone=CLIP ViT-L/14, venue&year=ICLR’242026.01 | 15.9 | 27.1 | 36.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ Qwen2.5-VLBackbone=OpenCLIP ViT-B/32, venue&year=reproduced2026.01 | 15.5 | 26.7 | 36.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ Qwen2.5-VLBackbone=OpenCLIP ViT-L/14, venue&year=reproduced2026.01 | 15.5 | 27 | 36.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ Qwen2.5-VLBackbone=CLIP ViT-L/14, venue&year=reproduced2026.01 | 15 | 26.4 | 36.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ GPT-4oBackbone=OpenCLIP ViT-L/14, venue&year=reproduced2026.01 | 15 | 25.9 | 35.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ Qwen2.5-VLBackbone=CLIP ViT-B/32, venue&year=reproduced2026.01 | 14.9 | 26.1 | 35.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SEARLEBackbone=CLIP ViT-B/32, venue&year=ICCV’232026.01 | 14.4 | 25.3 | 35.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ GPT-4oBackbone=CLIP ViT-L/14, venue&year=reproduced2026.01 | 14.4 | 25 | 34.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ GPT-4oBackbone=OpenCLIP ViT-B/32, venue&year=reproduced2026.01 | 14.2 | 25.9 | 36.5 | — | — | — | — | — | — | — | — | — | — | — | — | |
| OSrCIR w/ GPT-4oBackbone=CLIP ViT-B/32, venue&year=reproduced2026.01 | 14 | 25.4 | 34.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LinCIR + RTDBackbone=ViT-L2024.06 | 13.18 | 23.12 | 32.77 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LinCIR + RTDBackbone=ViT-B2024.06 | 12.83 | 22.83 | 32.22 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SEARLE + RTDBackbone=ViT-B2024.06 | 12.82 | 22.97 | 32.44 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Image+TextBackbone=CLIP ViT-B/32, venue&year=baseline2026.01 | 12.7 | 23.7 | 32.9 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CIG + LinCIRBackbone=CLIP ViT-L/14, venue&year=CVPR’252026.01 | 12.5 | 22.7 | 32.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LinCIRBackbone=ViT-L2024.06 | 12.45 | 22.66 | 32.06 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SEARLE + RTDBackbone=ViT-L2024.06 | 12.4 | 22.82 | 32.37 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SEARLEBackbone=ViT-L2024.06 | 12.3 | 22.08 | 31.29 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SEARLEBackbone=CLIP ViT-L/14, venue&year=ICCV’232026.01 | 12.3 | 22.1 | 31.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LinCIRBackbone=ViT-B2024.06 | 12.23 | 21.29 | 30.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LinCIRBackbone=CLIP ViT-L/14, venue&year=CVPR’242026.01 | 12.2 | 22.8 | 32.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SEARLEBackbone=ViT-B2024.06 | 12.19 | 22.56 | 32.03 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pic2Word + RTDBackbone=ViT-B2024.06 | 12.03 | 21.61 | 31.09 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pic2Word + RTDBackbone=ViT-L2024.06 | 11.92 | 22.32 | 31.33 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Image-onlyBackbone=CLIP ViT-B/32, venue&year=baseline2026.01 | 11.5 | 21.4 | 31.1 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pic2WordBackbone=CLIP ViT-L/14, venue&year=CVPR’232026.01 | 11.2 | 21.5 | 30.4 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pic2WordBackbone=ViT-L2024.06 | 11.18 | 21.45 | 30.55 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pic2WordBackbone=ViT-B2024.06 | 11.13 | 21.08 | 31.05 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Text-onlyBackbone=CLIP ViT-B/32, venue&year=baseline2026.01 | 8.4 | 17.1 | 25.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| LinCIRBackbone=CLIP L/142023.11 | — | — | — | 16.55 | 28.97 | 39.15 | 7.84 | 16.56 | 25.61 | 12.59 | 23.68 | 33.84 | 11.8 | 21.85 | 30.92 | |
| OursBackbone=CLIP L/142023.11 | — | — | — | 17.74 | 29.77 | 39.31 | 12.02 | 21.38 | 31.63 | 16.7 | 27.24 | 36.96 | 13.06 | 23.91 | 33.98 | |
| Pic2WordBackbone=CLIP L/142023.11 | — | — | — | 14.76 | 26.42 | 35.85 | 7.55 | 16.53 | 24.9 | 12.04 | 23.09 | 32.21 | 10.28 | 19.86 | 28.54 | |
| SEARLEBackbone=CLIP L/142023.11 | — | — | — | 16.69 | 27.47 | 37.42 | 7.84 | 16.76 | 25.18 | 12.38 | 23.17 | 33.01 | 12.15 | 21.06 | 29.6 |