Visual Document Retrieval on DocVQA
86.5NDCG@5ReAlign
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ReAlignRetriever Type=Multi-modal, Backbone=Qwen2026.04 | 86.5 | 87.4 | |
| ReAlignRetriever Type=Multi-modal, Backbone=Phi3V2026.04 | 80 | 81.7 | |
| NV-EmbedRetriever Type=Text-based2026.04 | 76.6 | 78.4 | |
| BM25Retriever Type=Text-based2026.04 | 75.6 | 76.7 | |
| VDocRetrieverRetriever Type=Multi-modal2026.04 | 75.2 | 76.9 | |
| E5-MistralRetriever Type=Text-based2026.04 | 71.8 | 73.7 | |
| BGERetriever Type=Text-based2026.04 | 70 | 71.9 | |
| DSERetriever Type=Multi-modal2026.04 | 69 | 70.5 | |
| E5-VRetriever Type=Multi-modal2026.04 | 62 | 63.9 | |
| SigLIP 2Retriever Type=Multi-modal2026.04 | 53.7 | 56.2 | |
| VLM2VecRetriever Type=Multi-modal2026.04 | 40.1 | 42.8 | |
| CLIPRetriever Type=Multi-modal2026.04 | 29.3 | 32.5 |