[2512.14735] PyFi: Toward Pyramid-like Financial Image Understanding for VLMs via Adversarial Agents

Quantitative Finance > Computational Finance arXiv:2512.14735 (q-fin) [Submitted on 11 Dec 2025 (v1), last revised 8 Apr 2026 (this version, v2)] Title:PyFi: Toward Pyramid-like Financial Image Understanding for VLMs via Adversarial Agents Authors:Yuqun Zhang, Yuxuan Zhao, Sijia Chen View a PDF of the paper titled PyFi: Toward Pyramid-like Financial Image Understanding for VLMs via Adversarial Agents, by Yuqun Zhang and 2 other authors View PDF HTML (experimental) Abstract:This paper proposes PyFi, a novel framework for pyramid-like financial image understanding that enables vision language models (VLMs) to reason through question chains in a progressive, simple-to-complex manner. At the core of PyFi is PyFi-600K, a dataset comprising 600K financial question-answer pairs organized into a reasoning pyramid: questions at the base require only basic perception, while those toward the apex demand increasing levels of capability in financial visual understanding and expertise. This data is scalable because it is synthesized without human annotations, using PyFi-adv, a multi-agent adversarial mechanism under the Monte Carlo Tree Search (MCTS) paradigm, in which, for each image, a challenger agent competes with a solver agent by generating question chains that progressively probe deeper capability levels in financial visual reasoning. Leveraging this dataset, we present fine-grained, hierarchical, and comprehensive evaluations of advanced VLMs in the financial domain. Moreover, f...