Research map / Vision-language and multimodal / Visual question answering
Overcoming Language Priors in Visual Question Answering with Adversarial Regularization
Sainandan Ramakrishnan, Aishwarya Agrawal, Stefan Lee
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 question-only adversarial regularization (ablation) | SAN · VQA-CP v2 test | overall accuracy: 24.96 -> 27.24 | Improved |
| Add difference-of-entropies regularization (ablation) | SAN · VQA-CP v2 test | overall accuracy: 24.96 -> 25.75 | Improved |
| Add both Q-Adv and DoE regularizers (ablation) | SAN · VQA-CP v2 test | overall accuracy: 24.96 -> 33.29 | Improved |
| Add question-only adversarial regularization (ablation) | UpDn · VQA-CP v2 test | overall accuracy: 39.74 -> 40.08 | Improved |
| Add difference-of-entropies regularization (ablation) | UpDn · VQA-CP v2 test | overall accuracy: 39.74 -> 40.43 | Improved |
| Add both Q-Adv and DoE regularizers (ablation) | UpDn · VQA-CP v2 test | overall accuracy: 39.74 -> 41.17 | Improved |
| Add both Q-Adv and DoE regularizers (ablation) | SAN · VQA-CP v1 test | overall accuracy: 26.88 -> 43.43 | Improved |
| Add both Q-Adv and DoE regularizers (ablation) | SAN · VQA v2 val | overall accuracy: 52.41 -> 52.31 | No clear effect |
| Add both Q-Adv and DoE regularizers (ablation) | UpDn · VQA v2 val | overall accuracy: 63.48 -> 62.75 | Hurt |
| Add both Q-Adv and DoE regularizers (ablation) | SAN · VQA v1 val | overall accuracy: 55.86 -> 52.15 | Hurt |
| Add Q-Adv and DoE to the question encoder (ablation) | SAN question-only model · VQA-CP v2 train | overall accuracy: 24.83 -> 13.84 | Hurt |
| Use DoE alongside Q-Adv while sweeping the Q-Adv coefficient (ablation) | SAN · VQA-CP v2 test | overall accuracy | Improved |
Closest papers by meaning
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.92)
- Human-Adversarial Visual Question Answering (NeurIPS 2021 · similarity 0.92)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.91)
- VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models (AAAI 2024 · similarity 0.91)
- Right this way: Can VLMs Guide Us to See More to Answer Questions? (NeurIPS 2024 · similarity 0.89)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 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.
