Research map / Vision-language and multimodal / Multimodal LLMs
V-REX: Benchmarking Exploratory Visual Reasoning via Chain-of-Questions
Chenrui Fan, Yijun Liang, Shweta Bhardwaj, Kwesi Cobbina, Ming Li, Tianyi Zhou
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 |
|---|---|---|---|
| provide ground-truth CoQ answers/questions as hints (Planning mode) | GPT-5 · V-REX final question | accuracy: 77.1 -> 86.0 | Improved |
| provide ground-truth CoQ questions (Following mode) | GPT-5 · V-REX final question | accuracy: 77.1 -> 81.4 | Improved |
| provide CoQ hints (Planning mode) | InternVL3-38B · V-REX final question | accuracy: 67.4 -> 72.0 | Improved |
| provide CoQ hints (Following mode) | InternVL3-38B · V-REX final question | accuracy: 67.4 -> 72.8 | Improved |
| provide CoQ Planning hints | Qwen2.5-VL-7B · V-REX final question (Deduction) | accuracy: 60.8 -> 35.2 | Hurt |
| provide CoQ Planning hints | InternVL3.5-1B · V-REX final question (overall) | accuracy: 43.3 -> 37.7 | Hurt |
| provide CoQ hints on Retrieval category | VLMs · V-REX Retrieval | accuracy | No clear effect |
| increase model size | VLMs on V-REX · Planning vs Following average | accuracy: 35.1 -> 84.4 | Improved |
| increase candidate questions from 4 to 8 | GPT-4o · V-REX Planning (average) | accuracy: 72.72 -> 41.06 | Hurt |
| increase candidate questions from 4 to 8 | InternVL3-38B · V-REX Planning (average) | accuracy: 81.57 -> 58.08 | Hurt |
| increase candidate questions from 4 to 8 | Qwen3-VL-2B-IT · V-REX Planning (average) | accuracy: 32.96 -> 14.25 | Hurt |
| remove image input (blindfold) | GPT-5 · V-REX Navigation | accuracy: 77.55 | No clear effect |
Closest papers by meaning
- Prism: A Framework for Decoupling and Assessing the Capabilities of VLMs (NeurIPS 2024 · similarity 0.91)
- Zebra-CoT: A Dataset for Interleaved Vision-Language Reasoning (ICLR 2026 · similarity 0.91)
- VisRes Bench: On Evaluating the Visual Reasoning Capabilities of VLMs (CVPR 2026 · similarity 0.91)
- Multi-step Visual Reasoning with Visual Tokens Scaling and Verification (NeurIPS 2025 · similarity 0.91)
- MiCo: Multi-image Contrast for Reinforcement Visual Reasoning (NeurIPS 2025 · similarity 0.91)
- VGR: Visual Grounded Reasoning (ICLR 2026 · similarity 0.91)
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.
