Research map / Vision-language and multimodal / Visual question answering
Unshuffling Data for Improved Generalization in Visual Question Answering
Damien Teney, Ehsan Abbasnejad, 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 variance regularizer across environment classifiers (ablation) | Multi-environment training · VQA-CP v2 test (Other) | accuracy: 39.14 -> 47.60 | Improved |
| use question-group environments instead of random environments (ablation) | Environment training · VQA-CP v2 test (Other) | accuracy: 43.51 -> 47.60 | Improved |
| use K-means clustered question environments (ablation) | Environment training · VQA-CP v2 test (Other) | accuracy: 43.33 -> 46.35 | Improved |
| use question-group instead of clustered environments (ablation) | Environment training · VQA-CP v2 test (Other) | accuracy: 46.35 -> 47.60 | Improved |
| use relative variance instead of absolute variance (ablation) | Variance regularizer · VQA-CP v2 test (Other) | accuracy: 46.17 -> 47.60 | Improved |
| add alternating optimization with 4 warm-up epochs (ablation) | Multi-environment training · VQA-CP v2 test (Other) | accuracy: 47.60 -> 48.06 | Improved |
| add alternating optimization with 6 warm-up epochs (ablation) | Multi-environment training · VQA-CP v2 test (Other) | accuracy: 47.60 -> 47.38 | No clear effect |
| replace random with environment training | Environments · VQA-CP v2 val (Other) | accuracy: 54.74 -> 53.87 | Mixed |
| apply method to full VQA-CP | Bottom-up top-down · VQA-CP v2 test overall | accuracy: 37.87 -> 42.39 | Improved |
| apply method to VQA v2 standard split | Bottom-up top-down · VQA v2 val | overall accuracy: 61.09 -> 61.08 | No clear effect |
| combine method with 4-model ensemble | Bottom-up top-down · VQA-CP v2 test overall | accuracy: 39.30 -> 43.37 | Improved |
| treat VQA v2 and VG as two environments rather than aggregating | Bottom-up top-down · VQA v2 val | overall accuracy: 63.32 -> 63.80 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- RUBi — RUBi: Reducing Unimodal Biases for Visual Question Answering (NeurIPS 2019) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Separating Skills and Concepts for Novel Visual Question Answering (CVPR 2021 · similarity 0.92)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.91)
- An Analysis of Visual Question Answering Algorithms (ICCV 2017 · similarity 0.91)
- Plenty is Plague: Fine-Grained Learning for Visual Question Answering (TPAMI 2019 · similarity 0.90)
- Greedy Gradient Ensemble for Robust Visual Question Answering (ICCV 2021 · similarity 0.90)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.89)
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.
