Research map / Vision-language and multimodal / Multimodal LLMs
PercepTax: Benchmarking Cross-Property Reasoning in Vision-Language Models
Jonathan Lee, Xingrui Wang, Jiawei Peng, Luoxin Ye, Zehan Zheng, Tiezheng Zhang, Tao Wang, Wufei Ma, Siyi Chen, Yu-Cheng Chou, Prakhar Kaushik, Alan Yuille
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 PercepTax taxonomy ICL exemplar | Gemini 2.5 Pro · Real cross-property | avg accuracy: 53.00 -> 54.86 | Improved |
| add PercepTax taxonomy ICL exemplar | GPT-5 · Real cross-property | avg accuracy: 39.84 -> 43.60 | Improved |
| add PercepTax taxonomy ICL exemplar | InternVL3.5-8B · Real cross-property | avg accuracy: 26.77 -> 32.89 | Improved |
| add PercepTax taxonomy ICL exemplar | Qwen3-VL-8B · Real cross-property | avg accuracy: 39.08 -> 40.19 | No clear effect |
| add PercepTax taxonomy ICL exemplar | Qwen3-VL-8B · Real cross-property (Material) | accuracy: 26.69 -> 25.66 | Hurt |
| add oracle ground-truth property labels (ablation) | Gemini 2.5 Pro · Real single-property | accuracy: 55.29 -> 82.05 | Improved |
| add oracle ground-truth property labels (ablation) | Gemini 2.5 Pro · Real cross-property | accuracy: 53.00 -> 62.56 | Improved |
| add oracle ground-truth property labels (ablation) | Gemini 2.5 Pro · Sim. cross-property | accuracy: 38.20 -> 43.33 | Improved |
| add oracle ground-truth property labels (ablation) | Qwen3-VL-32B · Sim. single-property | accuracy: 25.99 -> 29.41 | No clear effect |
| add oracle ground-truth property labels (ablation) | Qwen3-VL-32B · Real cross-property | accuracy: 41.66 -> 47.89 | Improved |
| cross-property task instead of object description | GPT-5 · Real Image Set | accuracy: 87.06 -> 39.84 | Hurt |
| cross-property task instead of object description | Gemini 2.5 Pro · Real Image Set | accuracy: 84.70 -> 53.00 | Hurt |
Closest papers by meaning
- Thinking in Structures: Evaluating Spatial Intelligence in Constraint-Governed Spaces (ICML 2026 · similarity 0.89)
- Pursuing Minimal Sufficiency in Spatial Reasoning (ICLR 2026 · similarity 0.88)
- Visual Structures Help Visual Reasoning: Addressing the Binding Problem in LVLMs (NeurIPS 2025 · similarity 0.87)
- Perception Tokens Enhance Visual Reasoning in Multimodal Language Models (CVPR 2025 · similarity 0.87)
- SceneCOT: Eliciting Chain-of-Thought Reasoning in 3D Scenes (ICLR 2026 · similarity 0.87)
- Perceive, Interact, Reason: Building Tool-Augmented Visual Agents for Spatial Reasoning (NeurIPS 2026 · similarity 0.87)
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.
