Research map / Vision-language and multimodal / Visual question answering
Inverse Visual Question Answering: A New Benchmark and VQA Diagnosis Tool
Feng Liu, Tao Xiang, Timothy M. Hospedales, Wankou Yang, Changyin Sun
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 the image (full iVQA model vs answer-only) (ablation) | iVQA question generator · iVQA test split (VQA/COCO, question generation) | ranking accuracy Acc@1 (%): 14.589 -> 30.814 | Improved |
| add the answer (full model vs image only) (ablation) | iVQA question generator · iVQA test split (VQA/COCO, question generation) | ranking accuracy Acc@1 (%): 13.012 -> 30.814 | Improved |
| add answer type hint to image-only VQG (ablation) | VQG (I) · iVQA test split (VQA/COCO, question generation) | ranking accuracy Acc@1 (%): 13.012 -> 20.277 | Improved |
| add actual answer instead of answer type (ablation) | I+AT · iVQA test split (VQA/COCO, question generation) | ranking accuracy Acc@1 (%): 20.277 -> 30.814 | Improved |
| re-rank VQG candidates with VQA conditioning (VQG+VQA) | VQG (I) · iVQA test split (VQA/COCO, question generation) | ranking accuracy Acc@1 (%): 13.012 -> 16.529 | Improved |
| use proposed variational iVQA model instead of adapted captioning SAT | SAT · iVQA test split (VQA/COCO, question generation) | CIDEr: 1.533 -> 1.682 | Improved |
| use proposed iVQA model instead of VQG+VQA | VQG+VQA · iVQA test split (VQA/COCO, question generation) | ranking accuracy Acc@1 (%): 16.529 -> 30.814 | Improved |
| add image to language-only baseline in iVQA | iVQA language bias baseline · iVQA test (bias gameability) | acc@1 (%): 14.59 -> 28.44 | Improved |
| add image to language-only baseline in VQA | VQA LSTM Q baseline · VQA test-dev (bias gameability) | accuracy (%): 48.76 -> 57.75 | Improved |
| train MLB-att on balanced VQA 2.0 instead of VQA 1.0 (MLB2-att) | MLB-att · VQA belief sets | correct beliefs (%): 58.8 -> 65.4 | Improved |
| train MLB-att on VQA 2.0 instead of VQA 1.0 | MLB-att · VQA 1.0 test-dev | VQA accuracy (%): 64.59 -> 62.19 | Hurt |
| add visual attention (MLB-att vs Vanilla) | Vanilla VQA · VQA belief sets | correct beliefs (%): 48.6 -> 58.8 | Improved |
Closest papers by meaning
- IVQA: Inverse Visual Question Answering (CVPR 2018 · similarity 0.96)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.94)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.94)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.94)
- UNK-VQA: A Dataset and a Probe Into the Abstention Ability of Multi-Modal Large Models (TPAMI 2024 · similarity 0.94)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.92)
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.
