Research map / Vision-language and multimodal / Visual question answering
Debiased Visual Question Answering from Feature and Sample Perspectives
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 negative samples with mismatched images (v~, q) (ablation) | UpDn · VQA-CP v2 test | accuracy (%): 41.53 -> 57.59 | Helped |
| add negative samples with mismatched questions (v, q~) (ablation) | UpDn · VQA-CP v2 test | accuracy (%): 41.53 -> 41.60 | No clear effect |
| combine both negative sample types (ablation) | UpDn · VQA-CP v2 test | accuracy (%): 57.59 -> 60.24 | Helped |
| add question-to-answer branch (ablation) | D-VQA · VQA-CP v2 test | accuracy (%): 60.24 -> 61.18 | Helped |
| add vision-to-answer branch (ablation) | D-VQA · VQA-CP v2 test | accuracy (%): 61.18 -> 61.91 | Helped |
| add language bias detection module (ablation) | D-VQA base · VQA-CP v2 test | accuracy (%): 59.93 -> 60.28 | Helped |
| add vision bias detection module (ablation) | D-VQA base · VQA-CP v2 test | accuracy (%): 59.93 -> 59.98 | No clear effect |
| add both bias detection modules (ablation) | D-VQA base · VQA-CP v2 test | accuracy (%): 59.93 -> 61.91 | Helped |
| tune lambda_q:lambda_v ratio to 1:0.7 (ablation) | D-VQA · VQA-CP v2 test | accuracy (%): 60.74 (1:0.1) -> 61.91 (1:0.7) | Helped |
| apply D-VQA to different backbone (LXMERT) | LXMERT · VQA-CP v2 test | accuracy (%): 46.23 -> 69.75 | Helped |
| apply D-VQA to SAN backbone | SAN · VQA-CP v2 test | accuracy (%): 39.11 -> 54.39 | Helped |
| D-VQA versus SSL-VQA with 20% training data | D-VQA · VQA-CP v2 test 20% data | accuracy (%): 52.71 -> 52.94 | Helped |
Closest papers by meaning
- Roses Are Red, Violets Are Blue... but Should VQA Expect Them To? (CVPR 2021 · similarity 0.95)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.94)
- Language‑Bias‑Resilient Visual Question Answering via Adaptive Multi‑Margin Collaborative Debiasing (NeurIPS 2025 · similarity 0.93)
- How Transferable Are Reasoning Patterns in VQA? (CVPR 2021 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.93)
- NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples (NeurIPS 2024 · similarity 0.92)
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.