Research map / Vision-language and multimodal / Visual question answering
Where to Look: Focus Regions for Visual Question Answering
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Select question-relevant image regions instead of using a whole-image representation | VQA question answering · VQA validation | accuracy (%): 57.83 -> 58.94 | Helped |
| Select question-relevant regions instead of uniformly averaging region features | VQA question answering · VQA validation | accuracy (%): 57.88 -> 58.94 | Helped |
| Add visual region selection to language-only scoring | VQA question answering · VQA validation | accuracy (%): 53.98 -> 58.94 | Helped |
| Use region selection rather than the Q+I baseline | VQA question answering · VQA Test-dev | accuracy (%): 58.97 -> 62.44 | Helped |
| Add parser-based semantic bins to the question representation (ablation) | VQA language model · VQA validation real-images subset | accuracy (%): 51.87 -> 53.98 | Helped |
| Use region selection instead of whole-image features | VQA question answering · VQA validation: what color | accuracy (%): 43.52 -> 53.96 | Helped |
| Use region selection instead of whole-image features | VQA question answering · VQA validation: identify scene (what room/sport) | accuracy (%): 76.65 -> 86.21 | Helped |
| Use region selection instead of whole-image features | VQA question answering · VQA validation: how many | accuracy (%): 36.84 -> 33.38 | Hurt |
| Compute relevance from concatenated vision and language features rather than an inner product in a shared latent space (ablation) | VQA region-selection model · VQA | language-dependent region weighting | Hurt |
| Use the full image rather than region selection | VQA question answering · VQA validation: how many | accuracy (%): 33.38 -> 36.84 | Helped |
Closest papers by meaning
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · similarity 0.92)
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.92)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 2025 · similarity 0.92)
- Answer-Type Prediction for Visual Question Answering (CVPR 2016 · similarity 0.92)
- Re-Attention for Visual Question Answering (AAAI 2020 · similarity 0.91)
- REVIVE: Regional Visual Representation Matters in Knowledge-Based Visual Question Answering (NeurIPS 2022 · similarity 0.91)
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.