Research map / Vision-language and multimodal / Visual question answering
Separating Skills and Concepts for Novel Visual Question Answering
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 |
|---|---|---|---|
| add skill matching and concept grounding contrastive losses | Base VQA model · VQA v2 novel skill-concept compositions | VQA accuracy: 62.82 -> 64.32 | Helped |
| add skill matching and concept grounding contrastive losses | Base VQA model · VQA v2 novel concepts | VQA accuracy: 57.78 -> 59.80 | Helped |
| add skill matching and concept grounding contrastive losses | Base VQA model · manually annotated concept grounding | Recall@5: 43.71 -> 59.12 | Helped |
| add skill matching and concept grounding contrastive losses | Base VQA model · VQA-CP | VQA performance: 40.98 -> 41.71 | Helped |
| add skill matching and concept grounding contrastive losses | Base VQA model · VQA v2 test-dev and test-std | VQA performance: 69.60 -> 69.78; 69.99 -> 70.09 | Helped |
| add skill matching loss only (ablation) | Base VQA model · VQA v2 average novel count compositions | VQA accuracy: 58.03 -> 58.83 | Helped |
| add skill matching loss only (ablation) | Base VQA model · VQA v2 average novel concepts | VQA accuracy: 60.42 -> 61.85 | Helped |
| add concept grounding loss only (ablation) | Base VQA model · VQA v2 average novel count compositions | VQA accuracy: 58.03 -> 59.80 | Helped |
| add concept grounding loss only (ablation) | Base VQA model · VQA v2 average novel concepts | VQA accuracy: 60.42 -> 62.06 | Helped |
| replace both contrastive losses with masked language modeling (ablation) | Base VQA model · VQA v2 average novel count compositions | VQA accuracy: 58.03 -> 58.41 | No clear effect |
| replace both contrastive losses with masked language modeling (ablation) | Base VQA model · VQA v2 average novel concepts | VQA accuracy: 60.42 -> 60.25 | Hurt |
| use random sampling instead of CCC reference selection (ablation) | concept grounding objective · VQA v2 novel concepts (lamp, fruit, fridge, surfer, flag) | accuracy change from Base: +2.80, +0.49, -0.16, -0.14, +0.50 | Mixed |
Closest papers by meaning
- SkillCLIP: Skill Aware Modality Fusion Visual Question Answering (Student Abstract) (AAAI 2024 · similarity 0.91)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.91)
- Knowledge Acquisition for Visual Question Answering via Iterative Querying (CVPR 2017 · similarity 0.91)
- Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answering (CVPR 2018 · similarity 0.91)
- Exploring the Effectiveness of Object-Centric Representations in Visual Question Answering: Comparative Insights with Foundation Models (ICLR 2025 · similarity 0.90)
- Object Attribute Matters in Visual Question Answering (AAAI 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.