Research map / Vision-language and multimodal / Visual question answering
OAD-Promoter: Enhancing Zero-Shot VQA Using Large Language Models with Object Attribute Description
Quanxing Xu, Ling Zhou, Feifei Zhang, Rubing Huang, Jinyu Tian
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 OEG module (ablation) | OAD-Promoter · OKVQA test (few-shot) | soft accuracy: 47.33 -> 54.68 | Improved |
| add MKA module (ablation) | OAD-Promoter · OKVQA test (few-shot) | soft accuracy: 47.33 -> 48.95 | Improved |
| add MKA on top of OEG (ablation) | OAD-Promoter · OKVQA test (few-shot) | soft accuracy: 54.68 -> 60.04 | Improved |
| add OEG module (ablation) | OAD-Promoter · OKVQA test (zero-shot) | soft accuracy: 42.50 -> 44.26 | Improved |
| add MKA module (ablation) | OAD-Promoter · OKVQA test (zero-shot) | soft accuracy: 42.50 -> 43.64 | Improved |
| add both OEG and MKA (ablation) | OAD-Promoter · OKVQA test (zero-shot) | soft accuracy: 42.50 -> 45.61 | Improved |
| increase initial memory examples K from 0 to 400 (ablation) | OAD-Promoter (MKA only) · OKVQA test (zero-shot) | soft accuracy: 43.64 -> 44.15 | Improved |
| use CQA-CQA-CQA instead of CCC-QAQAQA prompt layout (ablation) | OAD Prompt · OKVQA test (zero-shot) | soft accuracy: 44.82 -> 45.61 | Improved |
| integrate LMH debiasing into LLM VQA pipelines (ablation) | PICa, Prophet, GRACE · OKVQA test (few-shot) | soft accuracy: Prophet 61.11 -> 59.96 | Hurt |
| integrate LMH and CSS debiasing (ablation) | Prophet · VQA-CP (few-shot) | soft accuracy: 53.41 -> 55.31 | Improved |
| swap frozen LLM from OPT to GPT-3 | OAD-Promoter · OKVQA test (zero-shot) | soft accuracy: 45.58 -> 45.61 | No clear effect |
| use OAD-Promoter instead of Img2LLM with small/medium OPT | OPT-6.7B and 30B · OKVQA test (zero-shot) | soft accuracy: 38.20 -> 36.18 (6.7B); 41.82 -> 40.46 (30B) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- GRACE — GRACE: Graph-Based Contextual Debiasing for Fair Visual Question Answering (ECCV 2024) · better on 1 of 3 shared results
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
- Knowledge Exchange with Confidence: Cost-Effective LLM Integration for Reliable and Efficient Visual Question Answering (ICLR 2026 · similarity 0.90)
- Exploring the Effectiveness of Object-Centric Representations in Visual Question Answering: Comparative Insights with Foundation Models (ICLR 2025 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.90)
- Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models (ICLR 2024 · similarity 0.90)
- Object Attribute Matters in Visual Question Answering (AAAI 2024 · similarity 0.90)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based 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.
