Multi-view Composed Image Retrieval on Multi-dataset Average (DeepFashion, F200K, FashionGen) (val)
80.6R@5Ours
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OursParams=0.8B, Encoding=Joint2026.04 | 80.6 | 88.9 | 84.8 | |
| Qwen3-V-8BParams=8B, Encoding=Joint2026.04 | 78.6 | 86.3 | 82.5 | |
| Qwen3-V-8BParams=8B, Encoding=MaxSim2026.04 | 74.9 | 83.4 | 79.2 | |
| Qwen3-V-8BParams=8B, Encoding=MeanPool2026.04 | 73.8 | 82.9 | 78.4 | |
| Doubao-E-VParams=–, Encoding=MaxSim2026.04 | 70.6 | 80.2 | 75.4 | |
| Doubao-E-VParams=–, Encoding=MeanPool2026.04 | 68.3 | 78.3 | 73.3 | |
| Qwen3-V-2BParams=2B, Encoding=Joint2026.04 | 66.6 | 77.6 | 72.1 | |
| Qwen3-V-2BParams=2B, Encoding=MeanPool2026.04 | 64.5 | 75.9 | 70.2 | |
| Qwen3-V-2BParams=2B, Encoding=MaxSim2026.04 | 63.1 | 75.2 | 69.2 | |
| Doubao-E-VParams=–, Encoding=Joint2026.04 | 58 | 72.8 | 65.4 | |
| SPRCParams=1.2B, Encoding=MaxSim2026.04 | 44.3 | 54.8 | 49.6 | |
| SPRCParams=1.2B, Encoding=MeanPool2026.04 | 41.6 | 52 | 46.8 | |
| RzenEmbedParams=8B, Encoding=MeanPool2026.04 | 36 | 45.6 | 40.8 | |
| RzenEmbedParams=8B, Encoding=MaxSim2026.04 | 22.6 | 29.8 | 26.2 | |
| CLIP4CIRParams=0.25B, Encoding=MeanPool2026.04 | 19.2 | 27.4 | 23.3 | |
| RzenEmbedParams=8B, Encoding=Joint2026.04 | 18.6 | 24.9 | 21.8 | |
| CLIP4CIRParams=0.25B, Encoding=MaxSim2026.04 | 18.2 | 26.4 | 22.3 |