Local Image Retrieval (Text-to-Image) on SORCE-1K
71.7Recall@1ObjEmbed-4B
Evaluation results
| Method | Links | |
|---|---|---|
| ObjEmbed-4BModel Parameters=4B2026.02 | 71.7 | |
| ObjEmbed-2BModel Parameters=2B2026.02 | 67.3 | |
| Qwen3-VL-Embedding-8BModel Parameters=8B2026.02 | 49.1 | |
| QQMM-embed-v2-7BModel Parameters=v2-7B2026.02 | 47.7 | |
| FG-CLIP2Backbone=ViT-So/162026.02 | 46.9 | |
| Qwen3-VL-Embedding-2BModel Parameters=2B2026.02 | 45.4 | |
| UME-R1-7BModel Parameters=7B2026.02 | 41.3 | |
| MetaCLIP2Backbone=ViT-H/142026.02 | 37.5 | |
| UME-R1-2BModel Parameters=2B2026.02 | 36.1 | |
| RzenEmbed-7BModel Parameters=7B2026.02 | 36 | |
| SigLIP2Backbone=ViT-So/162026.02 | 34.2 | |
| CLIPBackbone=ViT-L/142026.02 | 32.6 | |
| GME-7BModel Parameters=7B2026.02 | 30.6 | |
| FG-CLIP2Backbone=ViT-So/16, Evaluation Protocol=RoIAlign2026.02 | 28.4 | |
| GME-2BModel Parameters=2B2026.02 | 28.3 | |
| VLM2Vec-2BModel Parameters=2B2026.02 | 26.3 | |
| VLM2Vec-V2-2BModel Parameters=V2-2B2026.02 | 24.8 |