Research map / Vision-language and multimodal / Visual question answering
Graph-Structured Representations 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 |
|---|---|---|---|
| Use graph-structured scene and question representations with learned node matching | VQA on balanced scenes · Balanced scenes, complementary pairs | accuracy: 34.73 -> 39.1 | Helped |
| Use graph-structured scene and question representations | VQA on abstract scenes · Abstract scenes, multiple choice | overall VQA score: 71.18 -> 74.37 | Helped |
| Use graph-structured scene and question representations | VQA on abstract scenes · Abstract scenes, open-ended | overall VQA score: 69.73 -> 70.42 | Helped |
| Use syntactic dependency edges instead of only previous/next word edges (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 37.9 -> 39.1 | Helped |
| Use pretrained word embeddings instead of embeddings trained from scratch (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 33.8 -> 39.1 | Helped |
| Represent relative spatial relations with scene edge features (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 36.8 -> 39.1 | Helped |
| Enable recurrent graph processing for the question (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 37.1 -> 39.1 | Helped |
| Enable recurrent graph processing for the scene (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 37.0 -> 39.1 | Helped |
| Enable recurrent graph processing for both question and scene (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 35.7 -> 39.1 | Helped |
| Use one graph-processing iteration for the question rather than the full model's typical four (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 39.0 -> 39.1 | No clear effect |
| Use one graph-processing iteration for the scene (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 37.9 -> 39.1 | Hurt |
| Use one graph-processing iteration for both question and scene (ablation) | Graph VQA · Balanced scenes, complementary pairs | accuracy: 39.1 -> 39.1 | No clear effect |
Closest papers by meaning
- Visual Question Answering with Question Representation Update (QRU) (NeurIPS 2016 · similarity 0.93)
- Learning Conditioned Graph Structures for Interpretable Visual Question Answering (NeurIPS 2018 · similarity 0.93)
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.92)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.92)
- Self-Critical Reasoning for Robust Visual Question Answering (NeurIPS 2019 · similarity 0.92)
- Exploring the Effectiveness of Object-Centric Representations in Visual Question Answering: Comparative Insights with Foundation Models (ICLR 2025 · 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.