Research map / Vision-language and multimodal / Visual question answering
Co-Attending Free-Form Regions and Detections With Multi-Modal Multiplicative Feature Embedding 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 |
|---|---|---|---|
| Combine free-form region and detection-box attention with multiplicative feature embedding | Dual-MFA · VQA test-std, open-ended All | accuracy: 65.07 -> 66.09 | Helped |
| Combine free-form region and detection-box attention with multiplicative feature embedding | Dual-MFA · COCO-QA test, All | accuracy: 65.40 -> 66.49 | Helped |
| Use element-wise multiplication rather than addition for joint embedding of the three inputs (ablation) | MFA single-branch attention · VQA validation | accuracy: 56.34 -> 59.01 | Helped |
| Add L2 normalization to the joint feature embedding (ablation) | MFA single-branch attention · VQA validation | accuracy: 59.01 -> 59.18 | Helped |
| Use L2 normalization rather than unsigned power operation of 1/3 (ablation) | MFA single-branch attention · VQA validation | accuracy: 58.93 -> 59.18 | Helped |
| Fuse the two attention-branch outputs by addition rather than multiplication (ablation) | Dual-MFA · VQA validation | accuracy: 58.35 -> 59.82 | Helped |
| Fuse the two attention-branch outputs by addition rather than concatenation (ablation) | Dual-MFA · VQA validation | accuracy: 58.53 -> 59.82 | Helped |
| Add the free-form region attention branch to the detection-only model (ablation) | Dual-MFA · VQA validation | accuracy: 55.57 -> 59.82 | Helped |
| Add the detection-box attention branch to the free-form-only model (ablation) | Dual-MFA · VQA validation | accuracy: 59.18 -> 59.82 | Helped |
| Replace MLB embedding with multiplicative feature embedding in both branches (ablation) | Dual-MFA · VQA validation | accuracy: 59.07 -> 59.82 | Helped |
| Use MFA rather than QRU for detection-based single-branch attention (ablation) | MFA-D · VQA validation | accuracy: 53.99 -> 55.57 | Helped |
| Use MFA rather than MLB for detection-based single-branch attention (ablation) | MFA-D · VQA validation | accuracy: 54.89 -> 55.57 | Helped |
Closest papers by meaning
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.92)
- Tips and Tricks for Visual Question Answering: Learnings From the 2017 Challenge (CVPR 2018 · similarity 0.92)
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · similarity 0.92)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.91)
- Deep Modular Co-Attention Networks for Visual Question Answering (CVPR 2019 · similarity 0.91)
- FOCUS: Internal MLLM Representations for Efficient Fine-Grained Visual Question Answering (NeurIPS 2025 · similarity 0.89)
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.