Research map / Vision-language and multimodal / Visual question answering
Why Does a Visual Question Have Different Answers?
Nilavra Bhattacharya, Qing Li, Danna Gurari
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 predicted answers (A) as input cue to question and image (ablation) | Q+I model · VizWiz | mean average precision: 45.73 -> 50.02 | Improved |
| add predicted answers (A) as input cue to question and image (ablation) | Q+I model · VQA_2.0 | mean average precision: 43.16 -> 44.55 | Improved |
| add predicted answers (A) as input cue, spam category excluded (ablation) | Q+I model · VQA_2.0 (spam excluded) | mean average precision: 47.87 -> 48.05 | No clear effect |
| add predicted answers (A) as input cue, spam category excluded (ablation) | Q+I model · VizWiz (spam excluded) | mean average precision: 50.59 -> 55.18 | Improved |
| add image input to question input (ablation) | Q model · VizWiz | mean average precision: 40.5 -> 45.73 | Improved |
| add image input to question input (ablation) | Q model · VQA_2.0 | mean average precision: 43.47 -> 43.16 | No clear effect |
| fine-tune the VQA model on the answer-difference datasets (ablation) | Q+I+A · VizWiz | mean average precision: 50.02 -> 50.01 | No clear effect |
| fine-tune the VQA model on the answer-difference datasets (ablation) | Q+I+A · VQA_2.0 | mean average precision: 44.55 -> 44.46 | No clear effect |
| replace predicted answers with ground truth soft-target answers (ablation) | Q+I+A · VizWiz | mean average precision: 50.02 -> 50.68 | No clear effect |
| replace predicted answers with ground truth soft-target answers (ablation) | Q+I+A · VQA_2.0 | mean average precision: 44.55 -> 44.09 | No clear effect |
| use proposed Q+I+A model instead of QI-relevance baseline | QI-Relevance · VizWiz | mean average precision: 31.71 -> 50.02 | Improved |
| use proposed Q+I+A model instead of QI-relevance baseline | QI-Relevance · VQA_2.0 | mean average precision: 32.23 -> 44.55 | Improved |
Closest papers by meaning
- VQA Therapy: Exploring Answer Differences by Visually Grounding Answers (ICCV 2023 · similarity 0.91)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.90)
- Answer Them All! Toward Universal Visual Question Answering Models (CVPR 2019 · similarity 0.89)
- Answer-Type Prediction for Visual Question Answering (CVPR 2016 · similarity 0.88)
- Revisiting Visual Question Answering Baselines (ECCV 2016 · similarity 0.88)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · 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.
