Multimodal Retrieval (q^t -> e^i) on MBE 3.0 1.0 (test)
10.24R@1MOON3.0
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MOON3.0Dim.=256, Zero-shot=true2026.04 | 10.24 | 28.68 | 38.93 | |
| GME-Qwen2VLDim.=1536, Zero-shot=true2026.04 | 5.78 | 16.52 | 23.08 | |
| MOON2.0Dim.=256, Zero-shot=true2026.04 | 5.51 | 18.63 | 28.22 | |
| MOONDim.=256, Zero-shot=true2026.04 | 4.11 | 12.44 | 17.76 | |
| Qwen3-VL-2B (SFT)Dim.=256, Zero-shot=true2026.04 | 3.01 | 11.13 | 16.77 | |
| Qwen3-VL-EmbeddingDim.=2048, Zero-shot=true2026.04 | 2.65 | 7.1 | 9.68 | |
| InternVL3.5-2B (SFT)Dim.=256, Zero-shot=true2026.04 | 1.85 | 5.33 | 7.58 | |
| MM-EmbedDim.=4096, Zero-shot=true2026.04 | 1.4 | 4.62 | 7.15 | |
| SigLIP2Dim.=3072, Zero-shot=true2026.04 | 0.79 | 3.02 | 4.95 | |
| CASLIE-SDim.=3072, Zero-shot=true2026.04 | 0.76 | 2.65 | 4.12 | |
| FashionCLIPDim.=512, Zero-shot=true2026.04 | 0.47 | 1.42 | 2.06 | |
| InternVL3.5-2BDim.=2048, Zero-shot=true2026.04 | 0.05 | 0.19 | 0.34 | |
| Qwen3-VL-2BDim.=2048, Zero-shot=true2026.04 | 0.04 | 0.18 | 0.32 |