Research map / Vision-language and multimodal / Visual question answering
Visual Question Answering as a Meta Learning Task
Damien Teney, Anton van den Hengel
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 dynamic weights (ablation) | 2 static prototypes, L2 · VQA-Numbers | average answer recall: 31.48 -> 31.81 | Improved |
| add dynamic prototypes on top of dynamic weights (ablation) | dynamic-weights model · VQA-Numbers | average answer recall: 31.81 -> 32.32 | Improved |
| swap dot product for L2 distance (ablation) | 1 static prototype per answer · VQA-Numbers | average answer recall: 29.72 -> 27.80 | Hurt |
| swap dot product for L1 similarity (ablation) | 1 static prototype per answer · VQA-Numbers | average answer recall: 29.72 -> 29.97 | No clear effect |
| use two static prototypes per answer with L2 similarity (ablation) | dot-product baseline · VQA-Numbers | average answer recall: 29.72 -> 31.48 | Improved |
| use two static prototypes with L1 similarity (ablation) | 2 static prototypes dot product · VQA-Numbers | average answer recall: 30.28 -> 28.34 | Hurt |
| add dynamic prototypes only (ablation) | VQA v2 baseline (100% data) · VQA v2 val | answer recall: 15.8 -> 18.0 | Improved |
| add dynamic weights only (ablation) | VQA v2 baseline (100% data) · VQA v2 val | answer recall: 15.8 -> 16.3 | Improved |
| add dynamic weights and prototypes together (ablation) | VQA v2 baseline (50% data) · VQA v2 val | answer recall: 14.0 -> 15.5 | Improved |
| use proposed model instead of baseline | VQA v2 baseline (100% data) · VQA v2 val | question accuracy: 59.8 -> 59.9 | No clear effect |
| remove static prototypes, keep only dynamic (ablation) | dynamic prototypes model (100% data) · VQA v2 val | answer recall: 18.0 -> 4.29 | Hurt |
| train on only two of seven answers with full support set | proposed model · VQA-Numbers | average answer recall: 14.28 -> 17.05 | Improved |
Closest papers by meaning
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.95)
- Actively Seeking and Learning From Live Data (CVPR 2019 · similarity 0.94)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.93)
- Revisiting Visual Question Answering Baselines (ECCV 2016 · 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 0.92)
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.
