Research map / Vision-language and multimodal / Visual question answering
Learning to Generate Visual Questions with Noisy Supervision
Read the paperCheck 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 |
|---|---|---|---|
| Remove visual hints (ablation) | DH-GAN generator · VQA 2.0 val | B@4: 22.60 -> 21.87 | Hurt |
| Replace answer hints with only answer type (also remove visual hints) (ablation) | DH-GAN generator · VQA 2.0 val | B@4: 21.87 -> 18.63 | Hurt |
| Add GAN discriminator on top of the pretrained generator (ablation) | DH-GAN · VQA 2.0 val | B@4: 22.60 -> 23.71 | Helped |
| Remove visual hints sampling module (ablation) | DH-GAN · VQA 2.0 val | B@4: 23.71 -> 23.53 | Hurt |
| Remove visual hints sampling module (ablation) | DH-GAN · VQA 2.0 val | METEOR: 22.91 -> 22.87 | Hurt |
| Add GAN discriminator (DH-GAN vs generator only) in human evaluation (ablation) | DH-GAN · VQA2.0 human study | relevance: 3.86 -> 4 | Helped |
| Replace Radial with DH-GAN for question generation | DH-GAN · VQA 2.0 val | CIDEr: 162.92 -> 191.06 | Helped |
| Replace DH-VQG with DH-GAN (GAN noisy-hint refinement) | DH-GAN · COCO-QA val | CIDEr: 175.18 -> 186.65 | Helped |
| Replace DH-VQG with DH-GAN | DH-GAN · VQA 2.0 val | B@4: 22.43 -> 23.71 | Helped |
| Augment BUTD/BAN training with Radial-generated questions | VQA models · VQA2.0 data augmentation | classification accuracy | Hurt |
| Augment VQA training with DH-GAN-generated questions | BUTD and BAN · VQA2.0 data augmentation (30/50/70% data) | classification accuracy | Helped |
| Add VQG-generated questions to train zero-shot VQA (Radial) | BUTD · ZS-VQA VQA2.0 | accuracy: 0% -> 18.15% | Helped |
Closest papers by meaning
- Explicitly Guided Difficulty-Controllable Visual Question Generation (AAAI 2025 · similarity 0.90)
- ConVQG: Contrastive Visual Question Generation with Multimodal Guidance (AAAI 2024 · similarity 0.90)
- VQ-VA World: Towards High-Quality Visual Question-Visual Answering (CVPR 2026 · similarity 0.90)
- Visual Question Generation as Dual Task of Visual Question Answering (CVPR 2018 · similarity 0.90)
- VQA With No Questions-Answers Training (CVPR 2020 · similarity 0.89)
- Learning Answer Embeddings for Visual Question Answering (CVPR 2018 · 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.