Research map / Vision-language and multimodal / Visual question answering
SQuINTing at VQA Models: Introspecting VQA Models With Sub-Questions
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 |
|---|---|---|---|
| Fine-tune on VQA-introspect sub-questions | Pythia · VQA-introspect val | Consistency (%): 71.73 -> 76.20 | Helped |
| Fine-tune on VQA-introspect sub-questions | Pythia · VQA overall | Accuracy: 64.95 -> 64.60 | Hurt |
| Fine-tune on VQA-introspect sub-questions | Pythia · VQA Reasoning split | Accuracy: 69.61 -> 69.64 | Helped |
| Align main-question and sub-question attention while training on both answers | Pythia · VQA-introspect val | Consistency (%): 71.73 -> 76.84 | Helped |
| Align main-question and sub-question attention while training on both answers | Pythia · VQA-introspect val | Attention correlation: 0.71 -> 0.75 | Helped |
| Align main-question and sub-question attention while training on both answers | Pythia · VQA Reasoning split | Accuracy: 69.61 -> 69.88 | Helped |
| Align main-question and sub-question attention while training on both answers | Pythia · VQA overall | Accuracy: 64.95 -> 64.73 | Hurt |
| Use SQuINT instead of ordinary VQA-introspect fine-tuning | Pythia · VQA-introspect val | Attention correlation: 0.71 -> 0.75 | Helped |
| Use SQuINT instead of ordinary VQA-introspect fine-tuning | Pythia · VQA-introspect val | Consistency (balanced) (%): 74.58 -> 75.76 | Helped |
Closest papers by meaning
- Perception Matters: Detecting Perception Failures of VQA Models Using Metamorphic Testing (CVPR 2021 · similarity 0.95)
- GQA: A New Dataset for Real-World Visual Reasoning and Compositional Question Answering (CVPR 2019 · similarity 0.92)
- Roses Are Red, Violets Are Blue... but Should VQA Expect Them To? (CVPR 2021 · similarity 0.92)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.90)
- Prism: A Framework for Decoupling and Assessing the Capabilities of VLMs (NeurIPS 2024 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · 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.