Image-to-Text Retrieval on ShareGPT4V
97.8Recall@1Qwen3-VL-Embedding
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3-VL-EmbeddingParameters=2B2026.02 | 97.8 | |
| FG-CLIP2Backbone=ViT-So/162026.02 | 97.5 | |
| ObjEmbedParameters=4B2026.02 | 97.5 | |
| FG-CLIP 2Backbone=ViT-So/16, Evaluation Protocol=Zero-shot2025.10 | 97.5 | |
| FG-CLIPBackbone=ViT-L/14, Evaluation Protocol=Zero-shot2025.10 | 97.4 | |
| Long-CLIP + GMAILalignment=GMAIL2026.02 | 97.2 | |
| ObjEmbedParameters=2B2026.02 | 97.1 | |
| FG-CLIP 2Backbone=ViT-L/16, Evaluation Protocol=Zero-shot2025.10 | 96.9 | |
| FG-CLIPBackbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 96.7 | |
| FG-CLIP 2Backbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 95.8 | |
| Long-CLIPBackbone=ViT-L/14, Evaluation Protocol=Zero-shot2025.10 | 95.8 | |
| Long-CLIPBackbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 94.7 | |
| Long-CLIP2026.02 | 94.6 | |
| MetaCLIP2Backbone=ViT-H/142026.02 | 93.9 | |
| Meta CLIP 2Backbone=ViT-H/14, Evaluation Protocol=Zero-shot2025.10 | 93.9 | |
| UME-R1Parameters=2B2026.02 | 92.9 | |
| GMEParameters=2B2026.02 | 92.8 | |
| VLM2Vec-V2Parameters=2B2026.02 | 92.1 | |
| EVA-CLIPBackbone=ViT-L/142026.02 | 91.5 | |
| EVA-CLIPBackbone=ViT-L/14, Evaluation Protocol=Zero-shot2025.10 | 91.5 | |
| EVA-CLIPBackbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 90.5 | |
| GMEParameters=7B2026.02 | 89.3 | |
| CLIPBackbone=ViT-L/142026.02 | 86.5 | |
| CLIPBackbone=ViT-L/14, Evaluation Protocol=Zero-shot2025.10 | 86.5 | |
| CLIP + GMAILalignment=GMAIL2026.02 | 85.2 | |
| SigLIP 2Backbone=ViT-L/16, Evaluation Protocol=Zero-shot2025.10 | 85.1 | |
| SigLIP2Backbone=ViT-So/162026.02 | 78.6 | |
| SigLIP 2Backbone=ViT-So/16, Evaluation Protocol=Zero-shot2025.10 | 78.6 | |
| CLIP2026.02 | 78.2 | |
| CLIPBackbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 78.2 | |
| VLM2VecParameters=2B2026.02 | 77.7 | |
| FineCLIPBackbone=ViT-L/14, Evaluation Protocol=Zero-shot2025.10 | 73.4 | |
| FineCLIPBackbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 70.6 | |
| SigLIP 2Backbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 66 |