Research map / Vision-language and multimodal / Visual question answering
Adversarial VQA: A New Benchmark for Evaluating the Robustness of VQA Models
Linjie Li, Jie Lei, Zhe Gan, Jingjing Liu
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 R1 adversarial data to training (ablation) | UNITER-B · AVQA test | accuracy (%): 18.81 -> 20.47 | Improved |
| add R1 and R2 data to training (ablation) | UNITER-B · AVQA test | accuracy (%): 18.81 -> 22.62 | Improved |
| add all AVQA rounds to training (ablation) | UNITER-B · AVQA test | accuracy (%): 18.81 -> 24.10 | Improved |
| add AVQA to training (ablation) | UNITER-B · VQA v2 test-dev | accuracy (%): 72.70 -> 72.66 | No clear effect |
| train on AVQA only instead of VQA v2+VG (ablation) | UNITER-B · VQA v2 test-dev | accuracy (%): 72.70 -> 59.99 | Hurt |
| add image features to the model (versus language-only) (ablation) | UNITER-B · R1 test | accuracy (%): 17.82 -> 17.91 | Improved |
| train on combined data (with correctly answered examples) versus verified only (ablation) | UNITER-B · R1 test | accuracy (%): 25.63 -> 26.85 | No clear effect |
| scale UNITER-B to UNITER-L (ablation) | UNITER · AVQA test | accuracy (%): 24.10 -> 24.78 | Mixed |
| add random-image regularization (ablation) | BUTD · AVQA val | accuracy (%): 23.91 -> 23.79 | Hurt |
| add random-image regularization (ablation) | UNITER-B · AVQA val | accuracy (%): 25.04 -> 24.70 | Hurt |
| add random-image regularization (ablation) | BUTD · VQA-CP v2 test | accuracy (%): 40.62 -> 43.96 | Improved |
| use grid-feature end-to-end ClipBERT instead of region-feature UNITER-B (ablation) | VQA model · AVQA test | accuracy (%): 18.81 -> 21.16 | No clear effect |
Closest papers by meaning
- VQAttack: Transferable Adversarial Attacks on Visual Question Answering via Pre-trained Models (AAAI 2024 · similarity 0.92)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.90)
- NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples (NeurIPS 2024 · similarity 0.90)
- On the Robustness of Large Multimodal Models Against Image Adversarial Attacks (CVPR 2024 · similarity 0.90)
- Improving Automatic VQA Evaluation Using Large Language Models (AAAI 2024 · similarity 0.89)
- Open-ended VQA benchmarking of Vision-Language models by exploiting Classification datasets and their semantic hierarchy (ICLR 2024 · similarity 0.88)
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.
