Research map / Vision-language and multimodal / Visual question answering
UNK-VQA: A Dataset and a Probe Into the Abstention Ability of Multi-Modal Large Models
Yangyang Guo, Fangkai Jiao, Zhiqi Shen, Liqiang Nie, Mohan Kankanhalli
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 post-hint explanation of why the question is unanswerable | Otter-MPT-7B · UNK-VQA (3-shot) | accuracy: 12.17 -> 67.39 | Improved |
| add post-hint explanation of why the question is unanswerable | Open-Flamingo-MPT-7B · UNK-VQA (3-shot) | accuracy: 54.79 -> 55.20 | Improved |
| add post-hint explanation of why the question is unanswerable | Open-Flamingo-MPT-7B · UNK-VQA (5-shot) | accuracy: 54.57 -> 57.75 | Improved |
| fine-tune LLaVA-1.6 Vicuna-13B on UNK-VQA | LLaVA-1.6 Vicuna-13B · UNK-VQA open-ended | accuracy: 21.53 -> 52.49 | Improved |
| fine-tune LLaVA-1.6 Mistral-7B on UNK-VQA | LLaVA-1.6 Mistral-7B · UNK-VQA binary | accuracy: 56.98 -> 56.35 | Hurt |
| swap Vicuna-13B LLM for stronger Mistral-7B | LLaVA-1.6 · UNK-VQA open-ended (zero-shot) | accuracy: 21.53 -> 49.30 | Improved |
| scale Instruct-BLIP Vicuna-7B to 13B | Instruct-BLIP · UNK-VQA zero-shot BY | accuracy: 43.98 -> 54.82 | Improved |
| scale Instruct-BLIP Vicuna-7B to 13B | Instruct-BLIP · UNK-VQA zero-shot MC | accuracy: 22.38 -> 18.41 | Hurt |
| use entropy-based selective function instead of classifier-based | LXMERT · UNK-VQA test | open accuracy: 19.45 -> 22.40 | Improved |
| use entropy-based selective function instead of classifier-based | UpDn · UNK-VQA test | open accuracy: 6.46 -> 5.91 | No clear effect |
| add 2-shot demonstrations | GPT-4V · UNK-VQA subset BY | accuracy: 58.0 -> 57.0 | No clear effect |
| add 2-shot demonstrations | GPT-4V · UNK-VQA subset MC | accuracy: 38.0 -> 43.0 | Improved |
Closest papers by meaning
- Inverse Visual Question Answering: A New Benchmark and VQA Diagnosis Tool (TPAMI 2018 · similarity 0.94)
- A-OKVQA: A Benchmark for Visual Question Answering Using World Knowledge (ECCV 2022 · similarity 0.93)
- Making the v in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering (CVPR 2017 · similarity 0.93)
- Toward Unsupervised Realistic Visual Question Answering (ICCV 2023 · similarity 0.93)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (CVPR 2025 · similarity 0.92)
- VTQA: Visual Text Question Answering via Entity Alignment and Cross-Media Reasoning (CVPR 2024 · similarity 0.92)
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.
