Research map / Vision-language and multimodal / Visual question answering
Rethinking Data Augmentation for Robust Visual Question Answering
Long Chen, Yuhang Zheng, Jun Xiao
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 KDDAug augmentation | UpDn · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 48.91 -> 61.52 | Improved |
| add KDDAug augmentation | UpDn · VQA-CP v2 test | accuracy: 39.74 -> 60.24 | Improved |
| add KDDAug augmentation | UpDn · VQA v2 val | accuracy: 63.48 -> 62.86 | Hurt |
| add KDDAug augmentation | LMH · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 57.24 -> 60.79 | Improved |
| add KDDAug augmentation | RUBi · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 49.66 -> 59.75 | Improved |
| add KDDAug augmentation | CSS+ · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 59.05 -> 61.65 | Improved |
| assign answers by KDDAug instead of SimpleAug on the same pairs (ablation) | UpDn+augmentation · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: over 8% on HM | Improved |
| use dynamic multi-teacher weights instead of OOD teacher only (ablation) | KDDAug · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 59.91 -> 61.52 | Improved |
| use dynamic multi-teacher weights instead of ID teacher only (ablation) | KDDAug · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 51.15 -> 61.52 | Improved |
| use dynamic multi-teacher weights instead of simple average (ablation) | KDDAug · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 57.79 -> 61.52 | Improved |
| add rule-based initial answers (ablation) | KDDAug · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 58.26 -> 61.52 | Improved |
| use more CLIP-filtered augmented samples (10% to 100%) (ablation) | KDDAug · VQA-CP v2 / VQA v2 val (HM) | harmonic mean: 60.85 -> 61.52 | Improved |
Closest papers by meaning
- Semantic Equivalent Adversarial Data Augmentation for Visual Question Answering (ECCV 2020 · similarity 0.92)
- Cycle-Consistency for Robust Visual Question Answering (CVPR 2019 · similarity 0.91)
- Plenty is Plague: Fine-Grained Learning for Visual Question Answering (TPAMI 2019 · similarity 0.91)
- Prompting Large Language Models With Answer Heuristics for Knowledge-Based Visual Question Answering (CVPR 2023 · similarity 0.91)
- Roses Are Red, Violets Are Blue... but Should VQA Expect Them To? (CVPR 2021 · 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.
