Research map / Vision-language and multimodal / Visual question answering
AQuA: Toward Strategic Response Generation for Ambiguous Visual Questions
Jihyoung Jang, Hyounghun Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add GRPO after SFT (ablation) | Qwen2.5-VL-3B-Tuned · AQuA test | overall strategic accuracy: 83.81 -> 86.28 | Improved |
| add GRPO after SFT (ablation) | InternVL3-2B-Tuned · AQuA test | overall strategic accuracy: 73.42 -> 79.11 | Improved |
| add GRPO after SFT (ablation) | Qwen2.5-VL-3B-Tuned · AQuA test (Level 2) | strategic accuracy: 61.33 -> 82.22 | Improved |
| add GRPO after SFT (ablation) | InternVL3-2B-Tuned · AQuA test (Level 2) | strategic accuracy: 37.67 -> 59.67 | Improved |
| add GRPO after SFT (ablation) | Qwen2.5-VL-3B-Tuned · AQuA test (Level 1) | strategic accuracy: 92.22 -> 77.0 | Hurt |
| add Chain-of-Thought prompt | Qwen2.5-VL-3B-Instruct · AQuA test | overall strategic accuracy: 32.83 -> 28.42 | Hurt |
| add Chain-of-Thought prompt | GPT-5 · AQuA test | overall strategic accuracy: 22.86 -> 25.72 | No clear effect |
| add strategy prompting | Qwen2.5-VL-3B-Instruct · AQuA test | overall strategic accuracy: 32.83 -> 25.42 | No clear effect |
| add strategy prompting | GPT-5 · AQuA test | overall strategic accuracy: 22.86 -> 42.25 | Improved |
| add strategy prompting | Qwen2.5-VL-72B-Instruct · AQuA test | overall strategic accuracy: 25.78 -> 42.22 | Improved |
| fine-tune on AQuA (SFT+GRPO) instead of zero-shot | Qwen2.5-VL-3B-Instruct · AQuA test | overall strategic accuracy: 32.83 -> 86.28 | Improved |
| fine-tune on AQuA (SFT+GRPO) instead of zero-shot | InternVL3-2B-Instruct · AQuA test | overall strategic accuracy: 25.95 -> 79.11 | Improved |
Closest papers by meaning
- Acknowledging Focus Ambiguity in Visual Questions (ICCV 2025 · similarity 0.92)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.91)
- Logical Implications for Visual Question Answering Consistency (CVPR 2023 · similarity 0.91)
- Knowledge Exchange with Confidence: Cost-Effective LLM Integration for Reliable and Efficient Visual Question Answering (ICLR 2026 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.91)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
