Research map / Vision-language and multimodal / Visual question answering
Semantic Equivalent Adversarial Data Augmentation for Visual Question Answering
Ruixue Tang, Chao Ma, Wei Emma Zhang, Qi Wu, Xiaokang Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 semantic equivalent adversarial data augmentation (visual + question) | BUTD · VQAv2 val | accuracy (%): 63.32 -> 65.16 | Improved |
| add question (paraphrase) augmentation (ablation) | BUTD + Ours · VQAv2 val | accuracy (%): 64.69 -> 65.16 | Improved |
| add visual adversarial augmentation (ablation) | BUTD + Ours · VQAv2 val | accuracy (%): 65.05 -> 65.16 | Improved |
| add EDA text-editing question augmentation | BUTD · VQAv2 val | accuracy (%): 63.32 -> 62.73 | Hurt |
| jointly train with random visual noise | BUTD · VQAv2 val | accuracy (%): 63.32 -> 63.28 | No clear effect |
| inject adversarial examples only in epochs (10,15) instead of (5,25) (ablation) | BUTD + Ours · VQAv2 val | accuracy (%): 63.93 -> 65.16 | Improved |
| inject adversarial examples only in epochs (10,15) instead of (15,20) (ablation) | BUTD + Ours · VQAv2 val | accuracy (%): 64.18 -> 65.16 | Improved |
| use augmentation with 20% of training data | BUTD · VQAv2 val (20% train) | accuracy (%): 55.45 -> 57.39 | Improved |
| use augmentation with 80% of training data | BUTD · VQAv2 val (80% train) | accuracy (%): 62.77 -> 64.27 | Improved |
| adversarial training against IFGSM test attack | BUTD · VQAv2 val (IFGSM attack) | accuracy (%): 30.83 -> 44.57 | Improved |
| adversarial training against PGD test attack | BUTD · VQAv2 val (PGD attack) | accuracy (%): 18.05 -> 22.64 | Improved |
| adversarial training against paraphrase attack | BUTD · VQAv2 val (paraphrase) | accuracy (%): 54.03 -> 63.18 | Improved |
Closest papers by meaning
- Human-Adversarial Visual Question Answering (NeurIPS 2021 · similarity 0.92)
- Rethinking Data Augmentation for Robust Visual Question Answering (ECCV 2022 · similarity 0.92)
- VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models (AAAI 2024 · similarity 0.92)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.91)
- Tips and Tricks for Visual Question Answering: Learnings From the 2017 Challenge (CVPR 2018 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 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.
