Research map / Vision-language and multimodal / Visual question answering
Goal-Oriented Visual Question Generation via Intermediate Rewards
Junjie Zhang, Qi Wu, Chunhua Shen, Jian Zhang, Jianfeng Lu, 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 progressive reward rp (ablation) | VQG-rg · GuessWhat?! NewObject (sampling) | game accuracy (%): 60.6 -> 62.1 | Improved |
| add informativeness reward ri (ablation) | VQG-rg · GuessWhat?! NewObject (sampling) | game accuracy (%): 60.6 -> 61.3 | Improved |
| add informativeness reward ri to rg+rp (ablation) | VQG-rg+rp · GuessWhat?! NewObject (sampling) | game accuracy (%): 62.1 -> 63.2 | Improved |
| add progressive reward rp to rg+ri (ablation) | VQG-rg+ri · GuessWhat?! NewImage (beam search) | game accuracy (%): 60.1 -> 60.8 | Improved |
| add goal-achieved reward with dialogue-round control over the supervised baseline | Baseline · GuessWhat?! NewObject (sampling) | game accuracy (%): 41.6 -> 60.6 | Improved |
| replace sole final-goal reward with three intermediate rewards | Sole-r · GuessWhat?! NewObject (sampling) | game accuracy (%): 58.5 -> 63.2 | Improved |
| replace sole final-goal reward with three intermediate rewards | Sole-r · GuessWhat?! NewImage (greedy) | game accuracy (%): 58.4 -> 60.7 | Improved |
| add progressive reward rp | Sole-r · GuessWhat?! progressive trend | successful games with progressive trend (%): 57.3 -> 60.7 | Improved |
| add informativeness reward ri | Sole-r · GuessWhat?! question informativeness | high-quality questions (%): 86.3 -> 87.7 | Improved |
| use intermediate rewards over Sole-r in human study | Sole-r · GuessWhat?! human study | human success rate (%): 69 -> 75 | Improved |
Closest papers by meaning
- Divide-and-Conquer: Tree-structured Strategy with Answer Distribution Estimator for Goal-Oriented Visual Dialogue (AAAI 2025 · similarity 0.89)
- Information Maximizing Visual Question Generation (CVPR 2019 · similarity 0.89)
- ConVQG: Contrastive Visual Question Generation with Multimodal Guidance (AAAI 2024 · similarity 0.88)
- TTRV: Test-Time Reinforcement Learning for Vision Language Models (CVPR 2026 · similarity 0.87)
- VisPlay: Self-Evolving Vision-Language Models (CVPR 2026 · similarity 0.87)
- Knowledge-based Visual Question Answer with Multimodal Processing, Retrieval and Filtering (NeurIPS 2025 · similarity 0.86)
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.
