AI Paper Atlas

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

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs

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 changedWhereMeasuredEffect
add PercepTax taxonomy ICL exemplarGemini 2.5 Pro · Real cross-propertyavg accuracy: 53.00 -> 54.86Improved
add PercepTax taxonomy ICL exemplarGPT-5 · Real cross-propertyavg accuracy: 39.84 -> 43.60Improved
add PercepTax taxonomy ICL exemplarInternVL3.5-8B · Real cross-propertyavg accuracy: 26.77 -> 32.89Improved
add PercepTax taxonomy ICL exemplarQwen3-VL-8B · Real cross-propertyavg accuracy: 39.08 -> 40.19No clear effect
add PercepTax taxonomy ICL exemplarQwen3-VL-8B · Real cross-property (Material)accuracy: 26.69 -> 25.66Hurt
add oracle ground-truth property labels (ablation)Gemini 2.5 Pro · Real single-propertyaccuracy: 55.29 -> 82.05Improved
add oracle ground-truth property labels (ablation)Gemini 2.5 Pro · Real cross-propertyaccuracy: 53.00 -> 62.56Improved
add oracle ground-truth property labels (ablation)Gemini 2.5 Pro · Sim. cross-propertyaccuracy: 38.20 -> 43.33Improved
add oracle ground-truth property labels (ablation)Qwen3-VL-32B · Sim. single-propertyaccuracy: 25.99 -> 29.41No clear effect
add oracle ground-truth property labels (ablation)Qwen3-VL-32B · Real cross-propertyaccuracy: 41.66 -> 47.89Improved
cross-property task instead of object descriptionGPT-5 · Real Image Setaccuracy: 87.06 -> 39.84Hurt
cross-property task instead of object descriptionGemini 2.5 Pro · Real Image Setaccuracy: 84.70 -> 53.00Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy