Research map / Vision-language and multimodal / Visual question answering
Counterfactual Samples Synthesizing for Robust Visual Question Answering
Check 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 V-CSS and Q-CSS | LMH · VQA-CP v2 | accuracy (All): 52.45 -> 58.95 | Helped |
| Use V-CSS only (ablation) | LMH · VQA-CP v2 | accuracy (All): 52.45 -> 58.23 | Helped |
| Use Q-CSS only (ablation) | LMH · VQA-CP v2 | accuracy (All): 52.45 -> 56.66 | Helped |
| Add V-CSS and Q-CSS | PoE · VQA-CP v2 | accuracy (All): 39.86 -> 48.32 | Helped |
| Use both V-CSS and Q-CSS instead of V-CSS only (ablation) | PoE · VQA-CP v2 | accuracy (All): 49.65 -> 48.32 | Hurt |
| Add V-CSS and Q-CSS | RUBi · VQA-CP v2 | accuracy (All): 45.23 -> 46.67 | Helped |
| Add V-CSS and Q-CSS | UpDn · VQA-CP v2 | accuracy (All): 39.68 -> 41.16 | Helped |
| Add CSS | LMH · VQA-CP v1 | accuracy (All): 55.27 -> 60.95 | Helped |
| Add CSS | LMH · VQA v2 val | accuracy (All): 61.64 -> 59.91 | Hurt |
| Use dynamic K for critical-object selection rather than fixed object counts (ablation) | LMH+V-CSS · VQA-CP v2 | accuracy | Helped |
| Replace only one critical word (ablation) | LMH+Q-CSS · VQA-CP v2 | accuracy | Helped |
| Set the V-CSS proportion to δ = 0.5 (ablation) | LMH+CSS · VQA-CP v2 | accuracy | Helped |
Closest papers by meaning
- Self-Critical Reasoning for Robust Visual Question Answering (NeurIPS 2019 · similarity 0.93)
- Roses Are Red, Violets Are Blue... but Should VQA Expect Them To? (CVPR 2021 · similarity 0.93)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.93)
- Debiased Visual Question Answering from Feature and Sample Perspectives (NeurIPS 2021 · similarity 0.92)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.91)
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.