Research map / Vision-language and multimodal / Visual question answering
Dynamic Fusion With Intra- and Inter-Modality Attention Flow 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 |
|---|---|---|---|
| Add both inter-modality and dynamically conditioned intra-modality attention flow (ablation) | DFAF · VQA 2.0 validation | Accuracy: 62.34 -> 66.21 | Helped |
| Add cross-modal conditioning to intra-modality attention (ablation) | DFAF · VQA 2.0 validation | Accuracy: 62.34 -> 65.51 | Helped |
| Add inter-modality attention flow (ablation) | DFAF · VQA 2.0 validation | Accuracy: 64.37 -> 66.21 | Helped |
| Stack additional DFAF blocks (ablation) | DFAF-1 · VQA 2.0 validation | Accuracy: 66.21 -> 66.66 | Helped |
| Use sequential rather than parallel InterMAF updates (ablation) | DFAF · VQA 2.0 validation | Accuracy: 65.99 -> 66.21 | Helped |
| Reverse the sequential InterMAF update order (ablation) | DFAF · VQA 2.0 validation | Accuracy: 66.21 -> 66.19 | No clear effect |
| Increase embedding dimension from 512 to 1024 (ablation) | DFAF · VQA 2.0 validation | Accuracy: 66.21 -> 65.89 | Hurt |
| Use element-wise multiplication rather than addition for final feature fusion (ablation) | DFAF · VQA 2.0 validation | Accuracy: 66.11 -> 66.21 | Helped |
| Use element-wise multiplication rather than concatenation for final feature fusion (ablation) | DFAF · VQA 2.0 validation | Accuracy: 66.14 -> 66.21 | Helped |
| Add visual sentinel features (ablation) | DFAF · VQA 2.0 validation | Accuracy: 66.21 -> 66.01; 66.21 -> 66.02 | Hurt |
| Add bounding-box position embeddings (ablation) | DFAF · VQA 2.0 validation | Accuracy: 66.21 -> 65.88; 66.21 -> 65.23 | Hurt |
| Use 8 attention heads instead of 1 or 4 (ablation) | DFAF · VQA 2.0 validation | Accuracy: 65.84 -> 66.21; 66.17 -> 66.21 | Helped |
Closest papers by meaning
- Cross-modal Information Flow in Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.88)
- Audio-Visual Adaptive Fusion Network for Question Answering Based on Contrastive Learning (AAAI 2025 · similarity 0.88)
- Object Attribute Matters in Visual Question Answering (AAAI 2024 · similarity 0.87)
- SkillCLIP: Skill Aware Modality Fusion Visual Question Answering (Student Abstract) (AAAI 2024 · similarity 0.87)
- Assessing Modality Bias in Video Question Answering Benchmarks with Multimodal Large Language Models (AAAI 2025 · similarity 0.87)
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.