Text-to-Image Retrieval on VGBench (repurposed)
35.03Recall@1Qwen2.5-VL-7B
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen2.5-VL-7BTraining=✗2026.04 | 35.03 | 61.77 | 68.46 | 71.37 | 74.56 | 46.34 | |
| LLaMA-3.2-11BTraining=✗2026.04 | 33.87 | 57.85 | 65.26 | 69.33 | 72.24 | 44.28 | |
| LLaVETraining=✓2026.04 | 23.26 | 33.43 | 35.61 | 37.06 | 37.35 | 27.57 | |
| VLM2VecTraining=✓2026.04 | 21.37 | 31.1 | 33.14 | 34.59 | 35.32 | 25.67 | |
| BLIPTraining=✗2026.04 | 20.78 | 30.81 | 33.87 | 35.47 | 36.48 | 25.21 | |
| CLIPTraining=✗2026.04 | 14.53 | 24.42 | 27.33 | 28.78 | 29.51 | 18.54 | |
| SigLIPTraining=✗2026.04 | 4.65 | 10.76 | 14.83 | 16.28 | 17.88 | 7.8 |