Research map / Vision-language and multimodal / Visual question answering
VQA-E: Explaining, Elaborating, and Enhancing Your Answers for Visual Questions
Qing Li, Qingyi Tao, Shafiq Joty, Jianfei Cai, Jiebo Luo
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 explanation generation as multi-task supervision (QI-AE vs QI-A) | VQA model with bottom-up features · VQA v2 val | accuracy (All): 61.78 -> 63.51 | Improved |
| add explanation supervision (ablation) | VQA model with grid features · VQA v2 val | accuracy (All): 59.25 -> 60.57 | Improved |
| add explanation supervision (ablation) | VQA model with global features · VQA v2 val | accuracy (All): 57.26 -> 57.92 | Improved |
| use relevant synthesized explanations instead of random captions as the explanation target (ablation) | QI-AE bottom-up · VQA v2 val | accuracy (All): 58.74 -> 63.51 | Improved |
| use random captions as explanation target instead of no explanation (ablation) | QI-AE bottom-up · VQA v2 val | accuracy (All): 61.78 -> 58.74 | Hurt |
| add answer prediction supervision to explanation model (QI-AE vs QI-E) | bottom-up features · VQA-E val | CIDEr-D: 84.07 -> 93.08 | Improved |
| switch from grid to bottom-up image features (ablation) | QI-AE · VQA-E val | CIDEr-D: 87.04 -> 93.08 | Improved |
| add attention (grid features) over global features (ablation) | QI-E · VQA-E val | CIDEr-D: 61.55 -> 73.70 | Improved |
| add the question as input to image-only explanation generation (ablation) | explanation model · VQA-E val | CIDEr-D: 48.58 -> 61.55 | Improved |
| use image input instead of question only for explanation (ablation) | explanation model · VQA-E val | CIDEr-D: 13.42 -> 48.58 | Improved |
| switch from grid to bottom-up image features (ablation) | QI-A · VQA v2 val | accuracy (All): 59.25 -> 61.78 | Improved |
| compare with BUTD on Number answers | QI-AE-Bottom-up · VQA v2 test-standard | accuracy (Number): 43.90 -> 43.58 | Hurt |
Closest papers by meaning
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.93)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.93)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.93)
- Self-Critical Reasoning for Robust Visual Question Answering (NeurIPS 2019 · similarity 0.93)
- Revisiting Visual Question Answering Baselines (ECCV 2016 · similarity 0.93)
- VQA: Visual Question Answering (IJCV 2016 · 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.
