Multimodal Question Answering on MultiModalQA 95 (test)
82.7AccuracyCogniVerse
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CogniVerseBackbone=LLaVA-13B, retrieval=hyperbolic embedding space2026.05 | 82.7 | 90 | 76.8 | 0.41 | |
| MMCoQAretrieval=sequence-level objectives2026.05 | 75.9 | 84 | 69.2 | 0.44 | |
| GraphRAGretrieval=static knowledge graphs2026.05 | 73.4 | 81 | 66.5 | 0.49 | |
| MuRAGretrieval=Euclidean embeddings2026.05 | 71.6 | 80 | 63.8 | 0.47 | |
| BLIP-2retrieval=none2026.05 | 65.2 | 76 | — | 0.2 | |
| CLIP-ViT-Lretrieval=none, fine-tuned=true2026.05 | 58.9 | 70 | — | 0.14 |