Research map / Vision-language and multimodal / Visual question answering
Deep Modular Co-Attention Networks 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 question self-attention (ablation) | MCAN with one MCA layer · VQA-v2 val | overall accuracy: 64.8 -> 65.2 | Helped |
| Add image self-attention (ablation) | MCAN with one MCA layer · VQA-v2 val | overall accuracy: 65.2 -> 65.4 | Helped |
| Use encoder-decoder instead of stacking with 6 MCA layers (ablation) | MCAN · VQA-v2 val | overall accuracy: 66.8 -> 67.2 | Helped |
| Increase encoder-decoder depth from 2 to 6 MCA layers (ablation) | MCAN · VQA-v2 val | overall accuracy: 66.2 -> 67.2 | Helped |
| Increase encoder-decoder depth from 6 to 8 MCA layers (ablation) | MCAN · VQA-v2 val | overall accuracy: 67.2 -> 67.2 | No clear effect |
| Use pretrained GloVe embeddings instead of randomly initialized embeddings (ablation) | MCAN_ed-6 question representation · VQA-v2 val | overall accuracy: 65.6 -> 67.0 | Helped |
| Use LSTM instead of positional encoding with pretrained GloVe (ablation) | MCAN_ed-6 question representation · VQA-v2 val | overall accuracy: 67.0 -> 67.1 | Helped |
| Fine-tune pretrained GloVe embeddings (ablation) | MCAN_ed-6 question representation · VQA-v2 val | overall accuracy: 67.1 -> 67.2 | Helped |
| Add image self-attention to the 6-layer model (ablation) | MCAN_ed-6 · VQA-v2 val, Number answers | accuracy: +4.5 points | Helped |
| Use MCAN_ed-6 instead of BAN | VQA-v2 test-dev · VQA-v2 test-dev | overall accuracy: +1.1 points | Helped |
| Use MCAN_ed-6 instead of BAN+Counter | VQA-v2 test-dev · VQA-v2 test-dev | overall accuracy: 0.6 points higher | Helped |
Closest papers by meaning
- Hierarchical Question-Image Co-Attention for Visual Question Answering (NeurIPS 2016 · similarity 0.94)
- Multi-Level Attention Networks for Visual Question Answering (CVPR 2017 · similarity 0.93)
- Stacked Attention Networks for Image Question Answering (CVPR 2016 · similarity 0.92)
- Improved Fusion of Visual and Language Representations by Dense Symmetric Co-Attention for Visual Question Answering (CVPR 2018 · similarity 0.92)
- Co-Attending Free-Form Regions and Detections With Multi-Modal Multiplicative Feature Embedding for Visual Question Answering (AAAI 2018 · similarity 0.91)
- Dynamic Capsule Attention for Visual Question Answering (AAAI 2019 · 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.