Research map / Vision-language and multimodal / Visual question answering
Counterfactual VQA: A Cause-Effect Look at Language Bias
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 |
|---|---|---|---|
| Apply CF-VQA SUM inference to UpDn (ablation) | UpDn · VQA-CP v2 test | All accuracy: 47.29 -> 53.55 | Helped |
| Apply CF-VQA SUM inference to S-MRL (ablation) | S-MRL · VQA-CP v2 test | All accuracy: 48.27 -> 55.05 | Helped |
| Apply CF-VQA HM inference to UpDn (ablation) | UpDn · VQA-CP v2 test | All accuracy: 47.97 -> 49.74 | Helped |
| Apply CF-VQA HM inference to S-MRL (ablation) | S-MRL · VQA-CP v2 test | All accuracy: 49.37 -> 51.27 | Helped |
| Replace RUBi's NIE inference with TIE inference (ablation) | RUBi with S-MRL · VQA-CP v2 test | All accuracy: 47.11 -> 54.69 | Helped |
| Use SUM instead of HM fusion (ablation) | CF-VQA with UpDn · VQA-CP v2 test | All accuracy: 49.74 -> 53.55 | Helped |
| Use prior-distributed rather than uniform no-treatment outputs (ablation) | CF-VQA (HM) with S-MRL · VQA-CP v2 test | All accuracy: 51.27 -> 46.29 | Hurt |
| Use unconstrained random rather than uniform no-treatment outputs (ablation) | CF-VQA (HM) with S-MRL · VQA-CP v2 test | All accuracy: 51.27 -> 31.27 | Hurt |
| Remove the direct vision-to-answer path in the causal graph (ablation) | CF-VQA (SUM) with S-MRL · VQA-CP v2 test | All accuracy: 55.05 -> 54.52 | No clear effect |
| Apply CF-VQA HM inference to UpDn (ablation) | UpDn · VQA-CP v2 test | Num. accuracy: 18.80 -> 18.46 | Hurt |
| Apply CF-VQA SUM inference to S-MRL (ablation) | S-MRL · VQA-CP v2 test | Other accuracy: 41.96 -> 45.61 | Hurt |
| Use CF-VQA inference on VQA v2 | UpDn · VQA v2 val | All accuracy: 63.48 -> 63.73 | No clear effect |
Closest papers by meaning
- What If the TV Was Off? Examining Counterfactual Reasoning Abilities of Multi-modal Language Models (CVPR 2024 · similarity 0.88)
- CF-VLM：CounterFactual Vision-Language Fine-tuning (NeurIPS 2025 · similarity 0.87)
- Scaling Test-Time Robustness of Vision-Language Models via Self-Critical Inference Framework (CVPR 2026 · similarity 0.87)
- Reasoning Elicitation in Language Models via Counterfactual Feedback (ICLR 2025 · similarity 0.86)
- Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation (ICML 2026 · similarity 0.86)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 2024 · similarity 0.86)
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.