Multi-view Composed Image Retrieval on FashionGen (val)
75Recall@5Ours
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OursParams=0.8B, Encoding=Joint2026.04 | 75 | 85.3 | |
| Qwen3-V-8BParams=8B, Encoding=Joint2026.04 | 74.7 | 83.5 | |
| Qwen3-V-8BParams=8B, Encoding=MaxSim2026.04 | 70.4 | 79.6 | |
| Qwen3-V-8BParams=8B, Encoding=MeanPool2026.04 | 67.9 | 78.4 | |
| Doubao-E-VParams=–, Encoding=MaxSim2026.04 | 67.2 | 77.1 | |
| Qwen3-V-2BParams=2B, Encoding=Joint2026.04 | 63 | 74.1 | |
| Doubao-E-VParams=–, Encoding=MeanPool2026.04 | 61.4 | 72.6 | |
| Qwen3-V-2BParams=2B, Encoding=MaxSim2026.04 | 58.2 | 69.9 | |
| Qwen3-V-2BParams=2B, Encoding=MeanPool2026.04 | 56.3 | 68.3 | |
| Doubao-E-VParams=–, Encoding=Joint2026.04 | 56.1 | 70.4 | |
| SPRCParams=1.2B, Encoding=MaxSim2026.04 | 42.7 | 53 | |
| SPRCParams=1.2B, Encoding=MeanPool2026.04 | 38.5 | 48.7 | |
| RzenEmbedParams=8B, Encoding=MeanPool2026.04 | 32.5 | 42.2 | |
| RzenEmbedParams=8B, Encoding=MaxSim2026.04 | 22.1 | 28.8 | |
| RzenEmbedParams=8B, Encoding=Joint2026.04 | 18.9 | 25.6 | |
| CLIP4CIRParams=0.25B, Encoding=MaxSim2026.04 | 17.1 | 25 | |
| CLIP4CIRParams=0.25B, Encoding=MeanPool2026.04 | 16.2 | 23.6 |