Research map / Vision-language and multimodal / Visual question answering
Learning by Asking Questions for Knowledge-Based Novel Object Recognition
K. Uehara, Tatsuya Harada
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 knowledge acquired by question generation (confidence policy) | CLIP-Ret. OC · Professional K-VQG merged (test set) | Overall acc. (zero-shot): 38.9 -> 56.1 | Improved |
| add knowledge acquired by question generation (confidence policy) | CLIP-Ret. OC · Professional K-VQG merged (test set) | Novel acc. (zero-shot): 0.0 -> 38.3 | Improved |
| fine-tune OC on acquired knowledge | Ours · Professional K-VQG merged (test set) | Overall acc.: 56.1 -> 62.0 | Improved |
| fine-tune OC on acquired knowledge | All Conf. · Professional K-VQG merged (test set) | Overall acc.: 56.2 -> 59.7 | Improved |
| fine-tune OC on acquired knowledge | All Exp. · Professional K-VQG merged (test set) | Known acc.: 57.1 -> 54.7 | Hurt |
| use mixed confirmation/exploration policy instead of all confirmation | Ours vs All Conf. · Professional K-VQG merged (test set) | Novel acc. (fine-tune): 45.3 -> 51.9 | Improved |
| use mixed confirmation/exploration policy instead of all confirmation | Ours vs All Conf. · Professional K-VQG merged (test set) | Overall acc. (zero-shot): 56.2 -> 56.1 | No clear effect |
| use confidence-based policy instead of random policy | Ours vs Random Policy · Professional K-VQG merged (test set) | Overall acc. (fine-tune): 53.2 ± 0.39 -> 62.0 | Improved |
| use confidence-based policy instead of all exploration | Ours vs All Exp. · Professional K-VQG merged (test set) | #valid questions: 1057 -> 9864 | Improved |
| use confidence-based policy instead of all exploration | Ours vs All Exp. · Professional K-VQG merged (test set) | Novel acc. (fine-tune): 28.4 -> 51.9 | Improved |
Closest papers by meaning
- Visual Question Generation for Class Acquisition of Unknown Objects (ECCV 2018 · similarity 0.91)
- An Empirical Evaluation of Visual Question Answering for Novel Objects (CVPR 2017 · similarity 0.89)
- Category-Guided Visual Question Generation (Student Abstract) (AAAI 2023 · similarity 0.85)
- VQA With No Questions-Answers Training (CVPR 2020 · similarity 0.85)
- Mining Object Parts From CNNs via Active Question-Answering (CVPR 2017 · similarity 0.85)
- Separation of Powers: On Segregating Knowledge from Observation in LLM-enabled Knowledge-based Visual Question Answering (CVPR 2025 · similarity 0.83)
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.
