Visual Question Answering on EVQA
40.8AccuracyPIXELRAG
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PIXELRAGReader=Qwen3-VL-4B, Top-k=3, Retrieval Category=Pixel-based retrieval2026.06 | 40.8 | 20.2 | |
| PIXELRAGReader=Qwen3-VL-4B, top-k=1, Retrieval Modality=Pixel-based retrieval2026.06 | 39.4 | 12.9 | |
| PIXELRAG (base)Reader=Qwen3-VL-4B, Top-k=3, Retrieval Category=Pixel-based retrieval2026.06 | 39.4 | 17 | |
| PIXELRAG (base)Reader=Qwen3-VL-4B, top-k=1, Retrieval Modality=Pixel-based retrieval2026.06 | 37 | 9.8 | |
| mwparserfromhellReader=Qwen3-VL-4B, Top-k=3, Retrieval Category=Text-based retrieval2026.06 | 30.7 | 3.2 | |
| mwparserfromhellReader=Qwen3-VL-4B, top-k=1, Retrieval Modality=Text-based retrieval2026.06 | 29.2 | 1.77 | |
| TrafilaturaReader=Qwen3-VL-4B, Top-k=3, Retrieval Category=Text-based retrieval2026.06 | 29.2 | 3.2 | |
| TrafilaturaReader=Qwen3-VL-4B, top-k=1, Retrieval Modality=Text-based retrieval2026.06 | 28 | 1.7 | |
| No retrievalReader=Qwen3-VL-4B, top-k=1, Retrieval Modality=None2026.06 | 27.3 | — | |
| No retrievalReader=Qwen3-VL-4B, Top-k=3, Retrieval Category=None2026.06 | 27.3 | — |