Image Query Multimodal Retrieval on M5Product (test)
32.8Recall@1VLM2Vec + AGCL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VLM2Vec + AGCLModel Architecture Category=LLM-based Baselines, Enhanced Component=AGCL2026.04 | 32.8 | 57 | 67.7 | |
| LLM2CLIP + AGCLModel Architecture Category=LLM-based Baselines, Enhanced Component=AGCL2026.04 | 32 | 58.6 | 69.7 | |
| VLM2VecModel Architecture Category=LLM-based Baselines2026.04 | 30.5 | 55.2 | 66.1 | |
| LLM2CLIPModel Architecture Category=LLM-based Baselines2026.04 | 30.1 | 56.8 | 68.2 | |
| BLIPModel Architecture Category=BERT-based Baselines2026.04 | 24.9 | 48.5 | 59.3 | |
| ALBEFModel Architecture Category=BERT-based Baselines2026.04 | 22.4 | 46.1 | 57.4 | |
| FG-CLIPModel Architecture Category=BERT-based Baselines2026.04 | 21.8 | 43.9 | 55.3 | |
| Region-CLIPModel Architecture Category=BERT-based Baselines2026.04 | 21.6 | 43.8 | 55.1 | |
| CLIPModel Architecture Category=BERT-based Baselines2026.04 | 21.5 | 43.6 | 54.6 | |
| VL-CLIPModel Architecture Category=BERT-based Baselines2026.04 | 21.5 | 43.8 | 54.9 | |
| MOON3.0Setting=Zero-shot2026.04 | 15.78 | 39.78 | 51.49 | |
| MOON2.0Setting=Zero-shot2026.04 | 11.28 | 24.43 | 32.37 | |
| Qwen3-VL-2B (SFT)Setting=SFT (Supervised Fine-Tuning)2026.04 | 10.6 | 23.94 | 34.25 | |
| MOONSetting=Zero-shot2026.04 | 9.11 | 21.59 | 26.53 | |
| FashionCLIPSetting=Zero-shot2026.04 | 8.23 | 19.44 | 26.13 | |
| SigLIP2Setting=Zero-shot2026.04 | 8.09 | 19.48 | 25.29 | |
| CASLIE-SSetting=Zero-shot2026.04 | 8 | 19.5 | 24.57 | |
| InternVL3.5-2B (SFT)Setting=SFT (Supervised Fine-Tuning)2026.04 | 7.92 | 19.31 | 25.66 | |
| GME-Qwen2VLSetting=Zero-shot2026.04 | 6.66 | 17.87 | 23.6 | |
| InternVL3.5-2BSetting=Zero-shot2026.04 | 5.83 | 15.16 | 21.02 | |
| MM-EmbedSetting=Zero-shot2026.04 | 5.78 | 13.57 | 17.14 | |
| Qwen3-VL-2BSetting=Zero-shot2026.04 | 5.38 | 13.1 | 16.91 | |
| Qwen3-VL-EmbeddingSetting=Zero-shot2026.04 | 4.51 | 10.67 | 14.25 |