Research map / Vision-language and multimodal / Visual question answering
Cycle-Consistency 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 |
|---|---|---|---|
| Train BAN with cycle-consistent question generation and answering | BAN · VQA-Rephrasings | CS(k=4): 39.87 -> 48.18 | Helped |
| Train BAN with cycle-consistent question generation and answering | BAN · VQA v2.0 test-dev | accuracy: 69.64 -> 69.87 | Helped |
| Add Q-consistency through question generation (ablation) | Pythia · VQA v2.0 val | accuracy: 65.78 -> 65.39 | No clear effect |
| Add A-consistency without filtering generated questions (ablation) | Pythia + Q-consistency · VQA v2.0 val | accuracy: 65.39 -> 62.08 | Hurt |
| Gate generated questions before applying A-consistency (ablation) | Pythia + A-consistency · VQA v2.0 val | accuracy: 62.08 -> 66.03 | Helped |
| Use unattended rather than VQA-attended image features for question generation (ablation) | Pythia + CC · VQA-Rephrasings | CS(k=4): 3.58% compared with 8.08% | Hurt |
| Use unattended rather than VQA-attended image features for question generation (ablation) | Pythia + CC · VQA v2.0 | accuracy: 0.2% compared with 0.5% | Hurt |
| Add an attention consistency loss matching attention across question rephrasings (ablation) | Pythia + CC · VQA v2.0 val | accuracy: 1.34% | Hurt |
| Train Pythia with cycle-consistent question generation and answering | Pythia · VQA-Rephrasings | CS(k=4): 39.49 -> 44.30 | Helped |
| Train BUTD with cycle-consistent question generation and answering | BUTD · VQA-Rephrasings | CS(k=4): 34.47 -> 42.55 | Helped |
| Train Pythia with cycle-consistent question generation and answering | Pythia · VQA v2.0 test-dev | accuracy: 68.43 -> 68.88 | Helped |
| Train BUTD with cycle-consistent question generation and answering | BUTD · VQA v2.0 test-dev | accuracy: 66.25 -> 67.55 | Helped |
Closest papers by meaning
- Logical Implications for Visual Question Answering Consistency (CVPR 2023 · similarity 0.94)
- Towards Causal VQA: Revealing and Reducing Spurious Correlations by Invariant and Covariant Semantic Editing (CVPR 2020 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.92)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.92)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.91)
- Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models (ICLR 2024 · 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.