Research map / Vision-language and multimodal / Visual question answering
An Empirical Evaluation of Visual Question Answering for Novel Objects
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Evaluate VQA models on questions containing novel objects rather than the original split | Architecture 1 · VQA open-ended overall | accuracy: 54.23 -> 39.38 | Hurt |
| Evaluate VQA models on questions containing novel objects rather than the original split | Architecture 2 · VQA open-ended overall | accuracy: 48.75 -> 34.97 | Hurt |
| Add text pretraining while keeping only the training vocabulary (ablation) | Architecture 1, Inception features · VQA novel split, open-ended overall | accuracy: 40.27 -> 40.18 | No clear effect |
| Add text pretraining while keeping only the training vocabulary (ablation) | Architecture 2, Inception features · VQA novel split, open-ended overall | accuracy: 37.66 -> 37.37 | No clear effect |
| Add novel words to the vocabulary and use text pretraining (ablation) | Architecture 1, Inception features · VQA novel split, open-ended overall | accuracy: 40.27 -> 41.19 | Helped |
| Add novel words to the vocabulary and use text pretraining (ablation) | Architecture 2, Inception features · VQA novel split, open-ended overall | accuracy: 37.66 -> 38.53 | Helped |
| Add vocabulary expansion to the general-setting text-pretrained model (ablation) | Architecture 1, Inception features · VQA novel split, open-ended overall | accuracy: 40.76 -> 41.39 | Helped |
| Add weakly paired image-text pretraining to text pretraining (ablation) | Architecture 1, Inception, oracle vocabulary · VQA novel split, open-ended overall | accuracy: 41.19 -> 40.73 | Hurt |
| Add weakly paired image-text pretraining to text pretraining (ablation) | Architecture 2, Inception, oracle vocabulary · VQA novel split, open-ended overall | accuracy: 38.53 -> 38.75 | Helped |
| Use general vocabulary expansion and auxiliary text training instead of the oracle vocabulary setting (ablation) | Architecture 1, early fusion · VQA novel split, open-ended overall | accuracy: 41.84 vs. 41.82 | No clear effect |
Closest papers by meaning
- Ask Me Anything: Free-Form Visual Question Answering Based on Knowledge From External Sources (CVPR 2016 · similarity 0.92)
- Answer-Type Prediction for Visual Question Answering (CVPR 2016 · similarity 0.92)
- REVIVE: Regional Visual Representation Matters in Knowledge-Based Visual Question Answering (NeurIPS 2022 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.90)
- Exploring the Effectiveness of Object-Centric Representations in Visual Question Answering: Comparative Insights with Foundation Models (ICLR 2025 · similarity 0.90)
- NaturalBench: Evaluating Vision-Language Models on Natural Adversarial Samples (NeurIPS 2024 · 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.