Multimodal Retrieval (q^i -> e^t) on MBE 3.0 1.0 (test)
11.39R@1MOON3.0
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MOON3.0Dim.=256, Zero-shot=true2026.04 | 11.39 | 31.98 | 42.95 | |
| MOON2.0Dim.=256, Zero-shot=true2026.04 | 6.62 | 21.9 | 32.47 | |
| GME-Qwen2VLDim.=1536, Zero-shot=true2026.04 | 5.13 | 15.84 | 22.71 | |
| MOONDim.=256, Zero-shot=true2026.04 | 4.51 | 13.62 | 19.3 | |
| Qwen3-VL-2B (SFT)Dim.=256, Zero-shot=true2026.04 | 3.31 | 10.06 | 17.12 | |
| Qwen3-VL-EmbeddingDim.=2048, Zero-shot=true2026.04 | 2.35 | 6.83 | 9.56 | |
| FashionCLIPDim.=512, Zero-shot=true2026.04 | 1.61 | — | 2.37 | |
| InternVL3.5-2B (SFT)Dim.=256, Zero-shot=true2026.04 | 1.58 | 4.89 | 7.07 | |
| SigLIP2Dim.=3072, Zero-shot=true2026.04 | 1.1 | 3.13 | 4.78 | |
| MM-EmbedDim.=4096, Zero-shot=true2026.04 | 1.01 | 3.56 | 5.66 | |
| CASLIE-SDim.=3072, Zero-shot=true2026.04 | 1 | 2.86 | 4.67 | |
| Qwen3-VL-2BDim.=2048, Zero-shot=true2026.04 | 0.34 | 1.35 | 2.27 | |
| InternVL3.5-2BDim.=2048, Zero-shot=true2026.04 | 0.09 | 0.36 | 0.65 |