Multi-view Composed Image Retrieval on F200K (val)
77.6R@5Ours
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OursParams=0.8B, Encoding=Joint2026.04 | 77.6 | 86.6 | |
| Qwen3-V-8BParams=8B, Encoding=Joint2026.04 | 73.8 | 82.1 | |
| Qwen3-V-8BParams=8B, Encoding=MaxSim2026.04 | 68.8 | 78.2 | |
| Qwen3-V-8BParams=8B, Encoding=MeanPool2026.04 | 68.1 | 77.8 | |
| Doubao-E-VParams=–, Encoding=MaxSim2026.04 | 62.2 | 72.6 | |
| Doubao-E-VParams=–, Encoding=MeanPool2026.04 | 61.2 | 71.8 | |
| Qwen3-V-2BParams=2B, Encoding=Joint2026.04 | 61.1 | 72.3 | |
| Qwen3-V-2BParams=2B, Encoding=MeanPool2026.04 | 60.3 | 72 | |
| Qwen3-V-2BParams=2B, Encoding=MaxSim2026.04 | 57.3 | 70.3 | |
| Doubao-E-VParams=–, Encoding=Joint2026.04 | 50.1 | 64 | |
| SPRCParams=1.2B, Encoding=MaxSim2026.04 | 34.4 | 43.7 | |
| SPRCParams=1.2B, Encoding=MeanPool2026.04 | 32.9 | 42.3 | |
| RzenEmbedParams=8B, Encoding=MeanPool2026.04 | 28 | 36.5 | |
| RzenEmbedParams=8B, Encoding=MaxSim2026.04 | 16 | 21.8 | |
| CLIP4CIRParams=0.25B, Encoding=MeanPool2026.04 | 13.3 | 19.3 | |
| RzenEmbedParams=8B, Encoding=Joint2026.04 | 12.7 | 17 | |
| CLIP4CIRParams=0.25B, Encoding=MaxSim2026.04 | 11.9 | 17.7 |