Research map / Vision-language and multimodal / Visual question answering
Robust Visual Question Answering: Datasets, Methods, and Future Challenges
Jie Ma, Pinghui Wang, Dechen Kong, Zewei Wang, Jun Liu, Hongbin Pei, Junzhou Zhao
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 |
|---|---|---|---|
| use pretrained LXMERT backbone with debiasing instead of UpDn | Loss-Rescaling · VQA-CP v2 test | accuracy: 47.09 -> 66.40 | Improved |
| add LMH ensemble debiasing to UpDn | UpDn · VQA v2 validation | accuracy: 65.99 -> 55.99 | Hurt |
| add CSS contrastive learning (CSS+CL) on LMH | LMH · VQA-CP v2 test | accuracy: 58.95 -> 59.18 | Mixed |
| generative ALBEF (4M) instead of discriminative | ALBEF · VQA-CP v2 test | accuracy: 44.40 -> 49.20 | Improved |
| scale ALBEF pretraining data from 4M to 14M | ALBEF DISC · VQA-CP v2 test | accuracy: 44.40 -> 45.20 | Improved |
| add large-scale pretraining to ALBEF GEN | ALBEF GEN · VQA-CP v2 test | accuracy: 36.60 -> 49.20 | Improved |
| move from in-distribution to OOD split | UNITER Base · VQA v2 val vs VQA-CP v2 | accuracy: 72.20 -> 46.93 | Hurt |
| use UNITER Large instead of Base | UNITER · VQA-CP v2 test | accuracy: 46.93 -> 50.98 | Improved |
| use SAR answer re-ranking on LMH | LMH · VQA-CP v2 test | accuracy: 52.87 -> 66.73 | Improved |
Closest papers by meaning
- Debiased Visual Question Answering from Feature and Sample Perspectives (NeurIPS 2021 · similarity 0.95)
- Cycle-Consistency for Robust Visual Question Answering (CVPR 2019 · similarity 0.94)
- Answer Them All! Toward Universal Visual Question Answering Models (CVPR 2019 · similarity 0.94)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.94)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.93)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.93)
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.
