Multimodal Question Answering on Encyclopedic-VQA 90 (test)
84.3AccuracyCogniVerse
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CogniVerseBackbone=LLaVA-13B, retrieval=hyperbolic embedding space2026.05 | 84.3 | 91 | 78.4 | 0.42 | |
| MMCoQAretrieval=sequence-level objectives2026.05 | 78.5 | 85 | 70.1 | 0.45 | |
| GraphRAGretrieval=static knowledge graphs2026.05 | 76.8 | 83 | 68.7 | 0.5 | |
| MuRAGretrieval=Euclidean embeddings2026.05 | 74.2 | 82 | 65.3 | 0.48 | |
| BLIP-2retrieval=none2026.05 | 68.7 | 78 | — | 0.22 | |
| CLIP-ViT-Lretrieval=none, fine-tuned=true2026.05 | 62.4 | 72 | — | 0.15 |