Research map / Vision-language and multimodal / Visual question answering
VQA: Visual Question Answering
Aishwarya Agrawal, Jiasen Lu, Stanislaw Antol, Margaret Mitchell, C. Lawrence Zitnick, Devi Parikh, Dhruv Batra
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 l2 normalization of image features (ablation) | deeper LSTM Q + norm I · VQA real images test-dev (open-ended) | accuracy: 57.59 -> 57.75 | Improved |
| fuse by element-wise multiplication instead of concatenation (ablation) | deeper LSTM Q + norm I · VQA real images test-dev (open-ended) | accuracy: 56.80 -> 57.75 | Improved |
| use 1000 answers instead of 500 (ablation) | deeper LSTM Q + norm I · VQA real images test-dev (open-ended) | accuracy: 56.93 -> 57.75 | Improved |
| use 2000 answers instead of 1000 | deeper LSTM Q + norm I · VQA real images test-dev (open-ended) | accuracy: 57.75 -> 58.15 | Improved |
| truncate question vocabulary at 5 occurrences | deeper LSTM Q + norm I · VQA real images test-dev (open-ended) | accuracy: 57.75 -> 57.99 | Improved |
| truncate question vocabulary at 11 occurrences | deeper LSTM Q + norm I · VQA real images test-dev (open-ended) | accuracy: 57.75 -> 57.81 | No clear effect |
| train on filtered dataset keeping confident answers | deeper LSTM Q + norm I · VQA real images test-dev (open-ended) | accuracy: 57.75 -> 56.62 | Hurt |
| add image features to question features (BoW Q + I) | BoW Q · VQA real images test-dev (open-ended) | accuracy: 48.09 -> 52.64 | Improved |
| add image to LSTM Q | LSTM Q · VQA real images test-dev (open-ended) | accuracy: 48.76 -> 53.74 | Improved |
| use a two-layer LSTM with normalized image features | LSTM Q + I · VQA real images test-dev (open-ended) | accuracy: 53.74 -> 57.75 | Improved |
| add human captions to BoW question | BoW Q · VQA real images val (open-ended) | accuracy: 48.09 -> 54.70 | Improved |
| add image features for reasoning-heavy question types (how many) | Question-only model · VQA question types (open-ended) | accuracy: 43.78 -> 40.33 | Hurt |
Closest papers by meaning
- Answer Them All! Toward Universal Visual Question Answering Models (CVPR 2019 · similarity 0.95)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.94)
- Visual Question Answering as Reading Comprehension (CVPR 2019 · similarity 0.94)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.94)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.94)
- Right this way: Can VLMs Guide Us to See More to Answer Questions? (NeurIPS 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.
