A benchmark and method for photographed table reasoning

Xiaoqiang Kang

doi:10.1016/j.patcog.2026.113355

A benchmark and method for photographed table reasoning

2026-02-25

0

PRE

AI

DOI：10.1016/j.patcog.2026.113355

Original

Original request for help

Save

Abstract

En 中文

• Introduce the first comprehensive study to address photographed table reasoning. • Release the first dataset for photographed Chinese table reasoning, PCTR-16K. • Propose Structure-aware Chain-of-Thought to boost multimodal table reasoning. • Integrate seven table structure understanding (TSU) tasks to enhance perception. • Conduct comprehensive experiments for various MLLMs with our proposed method.

Keywords:

photographed table reasoning

structured chain-of-thought

table structure understanding

multimodal reasoning

benchmark dataset

A benchmark and method for photographed table reasoning

Abstract

Journal

Organization