Research map / Vision-language and multimodal / Visual question answering
Stacked Attention Networks for Image 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 stacked visual attention rather than the strongest listed baseline | SAN on DAQUAR-ALL · DAQUAR-ALL | Accuracy: 23.4 -> 29.3 | Helped |
| use stacked visual attention rather than the strongest listed baseline | SAN on DAQUAR-REDUCED · DAQUAR-REDUCED | Accuracy: 39.7 -> 46.2 | Helped |
| use stacked visual attention rather than the strongest listed baseline | SAN on COCO-QA · COCO-QA | Accuracy: 55.9 -> 61.6 | Helped |
| use stacked visual attention rather than the strongest listed baseline | SAN on VQA · VQA test-dev | All: 53.7 -> 58.7 | Helped |
| add a second attention layer (ablation) | SAN with LSTM question model · DAQUAR-ALL | Accuracy: 28.9 -> 29.3 | Helped |
| add a second attention layer (ablation) | SAN with CNN question model · DAQUAR-ALL | Accuracy: 29.2 -> 29.3 | No clear effect |
| add a second attention layer (ablation) | SAN with LSTM question model · DAQUAR-REDUCED | Accuracy: 45.2 -> 46.2 | Helped |
| add a second attention layer (ablation) | SAN with CNN question model · COCO-QA | Accuracy: 60.7 -> 61.6 | Helped |
| add a second attention layer (ablation) | SAN with CNN question model · VQA local partition | All: 56.9 -> 57.6 | Helped |
| use three or more attention layers instead of two (ablation) | SAN · image QA benchmarks | performance | No clear effect |
| use a CNN rather than an LSTM question model (ablation) | SAN with two attention layers · DAQUAR-ALL | Accuracy: 29.3 vs 29.3 | No clear effect |
Closest papers by meaning
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.93)
- Deep Modular Co-Attention Networks for Visual Question Answering (CVPR 2019 · similarity 0.92)
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · similarity 0.91)
- Improved Fusion of Visual and Language Representations by Dense Symmetric Co-Attention for Visual Question Answering (CVPR 2018 · similarity 0.91)
- Re-Attention for Visual Question Answering (AAAI 2020 · similarity 0.90)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 2025 · similarity 0.88)
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.