Visual Question Answering and Grounding on VizWiz (test)
0.757CLIPScoreAutoVQA-G
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AutoVQA-G2026.04 | 0.757 | 0.8 | 0.874 | 64.9 | 68 | 0.737 | |
| GPT-4ochain-of-thought=true2026.04 | 0.754 | 0.841 | 0.901 | 35.4 | 34 | 0.59 | |
| GPT-4ozero-shot=true, tool-assisted=true2026.04 | 0.753 | 0.849 | 0.907 | 47.2 | 52.5 | 0.667 | |
| Geminizero-shot=true2026.04 | 0.745 | 0.818 | 0.883 | 44.5 | 48 | 0.639 | |
| Geminichain-of-thought=true2026.04 | 0.745 | 0.819 | 0.884 | 33.3 | 34 | 0.576 | |
| Human Annotation2026.04 | 0.724 | 0.794 | 0.762 | 62.7 | 64 | 0.697 |