Research map / Vision-language and multimodal / Visual question answering
REVIVE: Regional Visual Representation Matters in Knowledge-Based Visual Question Answering
Read the paperCheck 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 |
|---|---|---|---|
| use regional features instead of sliding-window features for knowledge retrieval (ablation) | REVIVE · OK-VQA | accuracy: 54.8 -> 56.6 | Helped |
| use regional features instead of whole-image features for retrieval (ablation) | REVIVE · OK-VQA | accuracy: 53.2 -> 56.6 | Helped |
| add positional coordinates of regions (ablation) | REVIVE · OK-VQA | accuracy: 55.8 -> 56.6 | Helped |
| add regional tags to GPT-3 prompt (implicit knowledge) (ablation) | REVIVE · OK-VQA | accuracy: 51.2 -> 52.4 | Helped |
| use regional features for explicit knowledge retrieval (ablation) | REVIVE · OK-VQA | accuracy: 52.9 -> 54.0 | Helped |
| add object-centric regional visual features to the encoder (ablation) | REVIVE · OK-VQA | accuracy: 54.0 -> 55.4 | Helped |
| add caption context to the question prompt (ablation) | REVIVE · OK-VQA | accuracy: 55.4 -> 55.9 | Helped |
| add regional tags to the context prompt (ablation) | REVIVE · OK-VQA | accuracy: 55.9 -> 56.6 | Helped |
| use GLIP detector instead of Faster R-CNN R101 (ablation) | REVIVE · OK-VQA | accuracy: 55.6 -> 56.6 | Helped |
| increase region proposals 36 -> 50 (ablation) | REVIVE · OK-VQA | accuracy: 56.6 -> 56.2 | Hurt |
| increase implicit candidates from 1 to 5 (ablation) | REVIVE · OK-VQA | accuracy: 55.8 -> 56.6 | Helped |
| increase implicit candidates 5 -> 8 (ablation) | REVIVE · OK-VQA | accuracy: 56.6 -> 56.4 | No clear effect |
Closest papers by meaning
- OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge (CVPR 2019 · similarity 0.94)
- Prompting Large Language Models With Answer Heuristics for Knowledge-Based Visual Question Answering (CVPR 2023 · similarity 0.94)
- KVQA: Knowledge-Aware Visual Question Answering (AAAI 2019 · similarity 0.93)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.92)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.92)
- BOK-VQA: Bilingual outside Knowledge-Based Visual Question Answering via Graph Representation Pretraining (AAAI 2024 · 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.