Research map / Vision-language and multimodal / Visual question answering
Towards Causal VQA: Revealing and Reducing Spurious Correlations by Invariant and Covariant Semantic Editing
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 IV-VQA invariant semantic edits to real training data | CNN+LSTM, SAAA, and SNMN · VQA v2 / IV-VQA | predictions flipped (%): CNN+LSTM 17.15→16.1; SAAA 7.53→7.09; SNMN 8.09→7.72 | Helped |
| Add IV-VQA invariant edits | CNN+LSTM · VQA v2 (all data) | accuracy (%): 60.21 →60.24 | No clear effect |
| Add IV-VQA invariant edits | SAAA · VQA v2 (all data) | accuracy (%): 70.25→69.25 | Hurt |
| Add IV-VQA invariant edits | SNMN · VQA v2 (all data) | accuracy (%): 67.65→68.02 | Helped |
| Add CV-VQA counting edits to real training data | CNN+LSTM, SAAA, and SNMN · CV-VQA counting test | predictions flipped (%): CNN+LSTM 83.84 → 50.74; SAAA 77.74→45.85; SNMN 73.12→44.19 | Helped |
| Add CV-VQA and IV-VQA edits to real training data | CNN+LSTM, SAAA, and SNMN · CV-VQA counting test | predictions flipped (%): CNN+LSTM 83.84 →59.58; SAAA 77.74→52.71; SNMN 77.13→51.91 | Helped |
| Add CV-VQA and IV-VQA edits to real training data | CNN+LSTM · CV-VQA counting test | accuracy (%): 43.65→43.94 | Helped |
| Add CV-VQA and IV-VQA edits to real training data | SAAA · CV-VQA counting test | accuracy (%): 50.87→50.45 | Hurt |
| Add CV-VQA and IV-VQA edits to real training data | SNMN · CV-VQA counting test | accuracy (%): 50.67→50.61 | No clear effect |
| Add IV-VQA edits for the 'is this a' question type | CNN+LSTM · IV-VQA, 'is this a' | predictions flipped (%): 12.72→9.77 | Helped |
| Add IV-VQA edits for the 'how many' question type | SAAA · IV-VQA, 'how many' | accuracy (%): 50.38→50.12 | Hurt |
Closest papers by meaning
- Cycle-Consistency for Robust Visual Question Answering (CVPR 2019 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.93)
- Logical Implications for Visual Question Answering Consistency (CVPR 2023 · similarity 0.93)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.92)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.91)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.90)
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.