Research map / Vision-language and multimodal / Visual question answering
Prophet: Prompting Large Language Models With Complementary Answer Heuristics for Knowledge-Based Visual Question Answering
Yu Zhou, Xuecheng Ouyang, Zhenwei Shao, Meng Wang, Jun Yu
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 answer candidates (K=1 vs none) (ablation) | Prophet · OK-VQA test | accuracy: 49.63 -> 56.04 | Improved |
| increase answer candidates from K=1 to K=10 (ablation) | Prophet · OK-VQA test | accuracy: 56.04 -> 60.84 | Improved |
| add answer-aware examples (fused features) instead of random examples (ablation) | Prophet · OK-VQA test | accuracy: 58.66 -> 60.84 | Improved |
| use fused features instead of PICa question+image selection (ablation) | Prophet · OK-VQA test | accuracy: 59.82 -> 60.84 | Improved |
| add question and image similarity to fused-feature selection (ablation) | Prophet · OK-VQA test | accuracy: 60.84 -> 60.38 | Hurt |
| select examples by answer logits instead of fused features (ablation) | Prophet · OK-VQA test | accuracy: 60.84 -> 60.40 | Hurt |
| add in-context examples (N=1 vs N=0, T=5) (ablation) | Prophet · OK-VQA test | accuracy: 49.97 -> 56.75 | Improved |
| increase queries from T=1 to T=5 at N=16 (ablation) | Prophet · OK-VQA test | accuracy: 57.52 -> 60.84 | Improved |
| add confidence scores to answer candidates (ablation) | Prophet · OK-VQA test | accuracy: 55.46 -> 60.84 | Improved |
| add image captions to prompt (ablation) | Prophet · OK-VQA test | accuracy: 58.27 -> 60.84 | Improved |
| add prompt head (ablation) | Prophet · OK-VQA test | accuracy: 60.54 -> 60.84 | No clear effect |
| add predicted object tags to prompt (ablation) | Prophet · OK-VQA test | accuracy: 60.84 -> 60.51 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PALI (17B) — PaLI: A Jointly-Scaled Multilingual Language-Image Model (ICLR 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Prompting Large Language Models With Answer Heuristics for Knowledge-Based Visual Question Answering (CVPR 2023 · similarity 0.97)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.93)
- Knowledge Exchange with Confidence: Cost-Effective LLM Integration for Reliable and Efficient Visual Question Answering (ICLR 2026 · similarity 0.93)
- Identifying and Resolving Pitfalls of Knowledge-Based VQA Benchmarks: Auditing, Repairing, and Augmenting (ECCV 2026 · similarity 0.93)
- TOA: Task-oriented Active VQA (NeurIPS 2023 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.92)
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.
