Research map / Vision-language and multimodal / Visual question answering
Acknowledging Focus Ambiguity in Visual Questions
Chongyan Chen, Yu-Yun Tseng, Zhuoheng Li, Anush Venkatesh, Danna Gurari
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 zero-shot chain-of-thought instruction (ablation) | Molmo-7B · VQ-FocusAmbiguity recognition | accuracy: 38.5 -> 56.9 | Improved |
| add enhanced chain-of-thought (structured steps) (ablation) | Qwen2.5-VL-7B · VQ-FocusAmbiguity recognition | accuracy: 57.2 -> 65.5 | Improved |
| add chain-of-thought prompting (ablation) | InternVL2-76B · VQ-FocusAmbiguity recognition | accuracy: 55.0 -> 56.7 | No clear effect |
| add chain-of-thought instruction (ablation) | GPT-4o · VQ-FocusAmbiguity recognition | accuracy: 67.7 -> 69.6 | Improved |
| add two few-shot examples (text-described) to zero-shot prompt (ablation) | GPT-4o · VQ-FocusAmbiguity recognition | accuracy: 67.7 -> 60.0 | Hurt |
| add two few-shot examples to zero-shot prompt (ablation) | Molmo-7B · VQ-FocusAmbiguity recognition | accuracy: 38.5 -> 41.8 | Improved |
| add ECoT step-by-step explanations to few-shot prompt (ablation) | Qwen2.5-VL-7B · VQ-FocusAmbiguity recognition | accuracy: 53.6 -> 59.0 | Improved |
| remove the image (question-only input) (ablation) | GPT-4o · VQ-FocusAmbiguity recognition | accuracy: 67.7 -> 44.2 | Hurt |
| replace image with self-generated caption at inference (few-shot) (ablation) | GPT-4o · VQ-FocusAmbiguity recognition | accuracy: +1.3% for FS; -2.1% for FS-ECoT | No clear effect |
| scale Qwen2.5-VL from 3B to 7B (ablation) | Qwen2.5-VL · VQ-FocusAmbiguity recognition | accuracy: 45.0 -> 65.5 | Improved |
| fine-tune on VQ-FocusAmbiguity (ablation) | Qwen2.5-VL-7B · VQ-FocusAmbiguity recognition | accuracy: +1.6 percentage points to 58.8% | Improved |
| add enhanced CoT to zero-shot prompt (ablation) | GLaMM · VQ-FocusAmbiguity localization | mAP: 13.01 -> 10.21 | Hurt |
Closest papers by meaning
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.92)
- AQuA: Toward Strategic Response Generation for Ambiguous Visual Questions (ICLR 2026 · similarity 0.92)
- Robust Visual Question Answering: Datasets, Methods, and Future Challenges (TPAMI 2024 · similarity 0.91)
- Logical Implications for Visual Question Answering Consistency (CVPR 2023 · similarity 0.91)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 2025 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 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.
