Research map / Vision-language and multimodal / Visual question answering
Attention and prior-robust VQA
Improves VQA via multi-level attention, bias reduction, decomposition, and cycle consistency.
109 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Visual question answering, keyword cluster “visual question · answering vqa · knowledge”
State History Tried Baselines Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
12 papers in 2023–24 and 32 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 10.8% to 18.5%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · 1,212 citations)
- 2017 · Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · 2,316 citations)
- 2018 · Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · 602 citations)
- 2019 · Dynamic Fusion With Intra- and Inter-Modality Attention Flow for Visual Question Answering (CVPR 2019 · 401 citations)
- 2020 · In Defense of Grid Features for Visual Question Answering (CVPR 2020 · 364 citations)
- 2021 · Benchmarking Knowledge-Enhanced Commonsense Question Answering via Knowledge-to-Text Transformation (AAAI 2021 · 42 citations)
- 2022 · MuKEA: Multimodal Knowledge Extraction and Accumulation for Knowledge-Based Visual Question Answering (CVPR 2022 · 128 citations)
- 2023 · RMLVQA: A Margin Loss Approach for Visual Question Answering With Language Biases (CVPR 2023 · 42 citations)
- 2024 · GRAM: Global Reasoning for Multi-Page VQA (CVPR 2024 · 13 citations)
- 2025 · Retrieval-Augmented Visual Question Answering via Built-in Autoregressive Search Engines (AAAI 2025 · 11 citations)
- 2026 · RoadSceneVQA: Benchmarking Visual Question Answering in Roadside Perception Systems for Intelligent Transportation System (AAAI 2026 · 2 citations)
Earliest papers here
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · 1,212 citations)
- Neural Module Networks (CVPR 2016 · 994 citations)
- Where to Look: Focus Regions for Visual Question Answering (CVPR 2016 · 538 citations)
What has been tried
84 papers here report 888 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 5%, hurt 17%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add 32 doc tokens rather than using no doc tokens | DUDE validation, all pages · ANLS | 46.16 -> 47.88 | Helped | GRAM: Global Reasoning for Multi-Page VQA (CVPR 2024 · 13 citations) |
| Add built-in autoregressive search engine | OKVQA · Score | 61.9 -> 65.7 | Helped | Retrieval-Augmented Visual Question Answering via Built-in Autoregressive Search Engines (AAAI 2025 · 11 citations) |
| Add the multimodal encoder features to image-encoder-only perturbation generation | VQAv2 and TextVQA transferable attacks (TCL source, VLMO-B target) · ASR | Helped | VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models (AAAI 2024 · 10 citations) | |
| add MGM during alignment training | DocVQA · score | 89.52 -> 92.01 | Helped | Marten: Visual Question Answering with Mask Generation for Multi-modal Document Understanding (CVPR 2025 · 7 citations) |
| Use ground-truth triples during training | BOK-VQA bilingual · accuracy | 45.08±0.94 -> 48.07±1.33 | Helped | BOK-VQA: Bilingual outside Knowledge-Based Visual Question Answering via Graph Representation Pretraining (AAAI 2024 · 5 citations) |
| Add QCIF, TKD-STFC, and QSCC | Split-AVQA · MA | 18.34 -> 32.05 | Helped | AVQACL: A Novel Benchmark for Audio-Visual Question Answering Continual Learning (CVPR 2025 · 2 citations) |
| Add QAFD to the baseline | OKVQA val · accuracy | 61.5 -> 62.9 | Helped | Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · 2 citations) |
| Use VQA-optimized hyperparameters rather than default hyperparameters | ImageNet-oVQA cropped · ClipM@1 | 57.10 -> 35.32 | Hurt | Open-ended VQA benchmarking of Vision-Language models by exploiting Classification datasets and their semantic hierarchy (ICLR 2024 · 2 citations) |
| add object descriptions and prompting for 2D object understanding | IaOR-VQA qualitative · Accuracy | 68.3 -> 87.3 | Helped | SpatialPIN: Enhancing Spatial Reasoning Capabilities of Vision-Language Models through Prompting and Interacting 3D Priors (NeurIPS 2024 · 2 citations) |
| Ensemble dense VQA models | VQA-v2 test · ECE | 0.14 -> 0.13 | Helped | GLEN: Generalized Focal Loss Ensemble of Low-Rank Networks for Calibrated Visual Question Answering (AAAI 2025 · 1 citation) |
| split the global VAE vector into K vectors | VQA downstream task · VQA performance | Helped | Exploring the Effectiveness of Object-Centric Representations in Visual Question Answering: Comparative Insights with Foundation Models (ICLR 2025 · 1 citation) | |
| Add reasoning context to MCTS | ScienceQA · accuracy | 88.92 -> 91.44 | Helped | Re-ranking Reasoning Context with Tree Search Makes Large Vision-Language Models Stronger (ICML 2025 · 0 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| TAP | ST-VQA test | ANLS | 0.543 – 0.597 | 2 papers, e.g. AAAI 2023 and CVPR 2022 |
What reviewers will expect
Datasets and baselines named in the 11 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- VQA-v2 · 2 of 11 papers
- ScienceQA · 2 of 11 papers
- OKVQA · 2 of 11 papers
- A-OKVQA · 2 of 11 papers
Baselines
- Prophet · 2 of 11 papers
- PromptCap · 2 of 11 papers
- REVIVE · 2 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 44 papers; from OpenAlex and the papers' affiliations).
- Amazon · 3 papers
- Beijing Institute of Technology · 3 papers
- Chinese Academy of Sciences · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Universal VQA models (100 papers)
- Vision-language model VQA evaluation (56 papers)
- Multimodal LLM question answering (56 papers)
Elsewhere in the topic
- Multimodal question answering datasets (90 papers)
- Open-domain multimodal QA (73 papers)
- Visual reasoning diagnostics (62 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).