Research map / Vision-language and multimodal / Visual question answering
Leveraging Visual Question Answering for Image-Caption Ranking
Xiao Lin, Devi Parikh
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 VQA-grounded score-level fusion to the VQA-agnostic model | Kiros et al. model · MSCOCO 1K test | caption retrieval R@1: 43.4 -> 46.9 | Improved |
| add VQA-grounded score-level fusion to the VQA-agnostic model | Kiros et al. model · MSCOCO 1K test | image retrieval R@1: 31.0 -> 35.8 | Improved |
| add VQA representations (N=3000) to deeper representation-level fusion model (ablation) | deeper VQA-agnostic model · MSCOCO 1K test | caption retrieval R@1: 45.8 -> 50.5 | Improved |
| add VQA representations (N=3000) to deeper representation-level fusion model (ablation) | deeper VQA-agnostic model · MSCOCO 1K test | image retrieval R@1: 33.6 -> 37.0 | Improved |
| add an additional layer to the VQA-agnostic model (ablation) | Kiros et al. model · MSCOCO 1K test | caption retrieval R@1: 43.4 -> 45.8 | Improved |
| add VQA representation for the caption only (ablation) | deeper VQA-agnostic model · MSCOCO 1K test | caption retrieval R@1: 45.8 -> 47.3 | Improved |
| add VQA representation for the image only (ablation) | deeper VQA-agnostic model · MSCOCO 1K test | image retrieval R@1: 33.6 -> 36.4 | Improved |
| add VQA representation for the image to caption-only fusion (ablation) | caption-only fusion model · MSCOCO 1K test | image retrieval R@1: 35.5 -> 37.0 | Improved |
| use VQA hidden activations z instead of VQA-grounded activations u (score-level fusion) (ablation) | score-level fusion VQA-aware · MSCOCO 1K test | caption retrieval R@1: 46.9 -> 46.8 | No clear effect |
| use VQA hidden activations z instead of VQA-grounded activations u (representation-level fusion) (ablation) | representation-level fusion VQA-aware · MSCOCO 1K test | caption retrieval R@1: 50.5 -> 49.3 | No clear effect |
| increase number of question-answer pairs N from 30 to 300 and beyond (ablation) | score-level and representation-level fusion · MSCOCO 1K test | retrieval recall | Mixed |
| sample 3 QA pairs per image on 1000 images instead of 1 QA pair per image on 3000 images (ablation) | N=3000 score-level fusion · MSCOCO 1K test | image/caption R@1: 46.9 -> 47.7 | No clear effect |
Closest papers by meaning
- VQA: Visual Question Answering (IJCV 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)
- VQA4CIR: Boosting Composed Image Retrieval with Visual Question Answering (AAAI 2025 · similarity 0.92)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.91)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.91)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.91)
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.
