Multi-view Composed Image Retrieval on DeepFashion (val)
89.2R@5Ours
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OursParams=0.8B, Encoding=Joint2026.04 | 89.2 | 94.9 | |
| Qwen3-V-8BParams=8B, Encoding=Joint2026.04 | 87.4 | 93.2 | |
| Qwen3-V-8BParams=8B, Encoding=MaxSim2026.04 | 85.6 | 92.3 | |
| Qwen3-V-8BParams=8B, Encoding=MeanPool2026.04 | 85.3 | 92.6 | |
| Doubao-E-VParams=–, Encoding=MeanPool2026.04 | 82.4 | 90.5 | |
| Doubao-E-VParams=–, Encoding=MaxSim2026.04 | 82.3 | 90.8 | |
| Qwen3-V-2BParams=2B, Encoding=MeanPool2026.04 | 76.8 | 87.5 | |
| Qwen3-V-2BParams=2B, Encoding=Joint2026.04 | 75.7 | 86.4 | |
| Qwen3-V-2BParams=2B, Encoding=MaxSim2026.04 | 73.9 | 85.4 | |
| Doubao-E-VParams=–, Encoding=Joint2026.04 | 67.8 | 84 | |
| SPRCParams=1.2B, Encoding=MaxSim2026.04 | 55.8 | 67.7 | |
| SPRCParams=1.2B, Encoding=MeanPool2026.04 | 53.4 | 65.1 | |
| RzenEmbedParams=8B, Encoding=MeanPool2026.04 | 47.5 | 58 | |
| RzenEmbedParams=8B, Encoding=MaxSim2026.04 | 29.6 | 38.8 | |
| CLIP4CIRParams=0.25B, Encoding=MeanPool2026.04 | 28 | 39.3 | |
| CLIP4CIRParams=0.25B, Encoding=MaxSim2026.04 | 25.7 | 36.6 | |
| RzenEmbedParams=8B, Encoding=Joint2026.04 | 24.3 | 32 |