Research map / Vision-language and multimodal / Visual question answering
Hierarchical Question-Image Co-Attention for Visual Question Answering
Read the paperCheck 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 |
|---|---|---|---|
| add question-level co-attention (vs uniform attention at question level) (ablation) | Ours^a+VGG hierarchical model · VQA validation | Overall accuracy: 57.0 vs 55.3 (+1.7) | Helped |
| add phrase-level co-attention (vs uniform) (ablation) | Ours^a+VGG hierarchical model · VQA validation | Overall accuracy: 57.0 vs 56.7 (+0.3) | Helped |
| add word-level co-attention (vs uniform) (ablation) | Ours^a+VGG hierarchical model · VQA validation | Overall accuracy: 57.0 vs 56.8 (+0.2) | Helped |
| use question attention in addition to image attention (full co-attention vs image attention alone) (ablation) | Ours^a+VGG · VQA validation | Overall accuracy: 57.0 vs 55.9 (+1.1) | Helped |
| question attention alone (no image attention) (ablation) | Ours^a+VGG · VQA validation | Overall accuracy: 54.8 vs 54.3 (+0.5) over LSTM Q+I; 54.8 vs 55.9 for image attention alone (-1.1) | Mixed |
| add image attention alone (vs holistic image feature) (ablation) | LSTM Q+I baseline · VQA validation | Overall accuracy: 55.9 vs 54.3 (+1.6) | Helped |
| switch VGG to ResNet features in alternating co-attention model | Ours^a · VQA test-dev open-ended | Overall accuracy: 61.8 vs 60.5 (+1.3) | Helped |
| switch VGG to ResNet features in alternating co-attention model | Ours^a on COCO-QA · COCO-QA test | Accuracy: 65.4 vs 62.9 (+2.5) | Helped |
| parallel vs alternating co-attention | COCO-QA · COCO-QA test | Accuracy: 63.3 (parallel) vs 62.9 (alternating) with VGG | Mixed |
| alternating co-attention vs parallel co-attention | VQA · VQA test-dev open-ended | Overall accuracy: 60.5 (alt) vs 60.1 (parallel) with VGG | Mixed |
Closest papers by meaning
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.94)
- Deep Modular Co-Attention Networks for Visual Question Answering (CVPR 2019 · similarity 0.94)
- Improved Fusion of Visual and Language Representations by Dense Symmetric Co-Attention for Visual Question Answering (CVPR 2018 · similarity 0.93)
- Re-Attention for Visual Question Answering (AAAI 2020 · similarity 0.93)
- Where to Look: Focus Regions for Visual Question Answering (CVPR 2016 · similarity 0.92)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 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.