Knowledge-based Visual Question Answering on Infoseek M2KR
50.5AccuracyPMSR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PMSRRetriever=SigLIP2-g, Model=Gemini-2.5-Flash, Final answer generator=LLaVA-MORE-8B (ReflectiVA)2025.08 | 50.5 | — | — | |
| PMSRRetriever=SigLIP2-g, Model=Qwen3-VL-8B, Final answer generator=LLaVA-MORE-8B (ReflectiVA)2025.08 | 41.5 | — | — | |
| PMSRRetriever=SigLIP2-g, Model=Qwen3-VL-4B, Final answer generator=LLaVA-MORE-8B (ReflectiVA)2025.08 | 38.3 | — | — | |
| MMKB-RAGRetriever=PreFLMR ViT-G, Model=Qwen2-VL-7B, Reranking=true2025.08 | 34.7 | — | — | |
| RA-VQAv2 w/ PreFLMRRetriever=PreFLMR2024.02 | 30.65 | — | — | |
| Baseline (no retrieval)Retrieval=None2024.02 | 21.8 | — | — | |
| RA-VQAv2 w/o retrievalRetrieval=None2024.02 | 21.78 | — | — | |
| AVISType=Hybrid system2024.02 | — | 50.7 | 56.4 | |
| PALI-XParameters=55B2024.02 | — | 50.7 | 56.4 |