跳到正文
arXiv cs.AI· Kecheng Liang, Haoyang Liu, Zexin Chen, Zirong Liu, Weixing Chen, Qiufeng Wang, Yang Liu, Liang Lin·· 2 小时前AI 评分36

PhysAlign:面向多模态物理推理中证据角色对齐的评测基准

PhysAlign: A Benchmark for Evidence-Grounded Role Alignment in Multimodal Physics Reasoning

AI 导读

PhysAlign 是一个评估多模态模型能否将物理图形中识别的信息正确对应到物理角色的基准,包含 3,341 个人工验证的探针,覆盖 986 道物理题,并引入 CAcc、GAcc、JAcc 等五项评测指标。评测显示,即使局部视觉识别正确,条件对应错误率在 GPT-6-Astra 上为 13.8%,InternVL3.5-8B 上约 50.6%,说明感知能力强不等于可靠的物理解释。

正文 · 原文

View PDF HTML (experimental)

Abstract:A key challenge in physics diagram understanding is correctly associating visual information with the physical entities, relations, and conditions it describes. Even when a value, symbol, or other local element is accurately recognized, assigning it to the wrong entity or scope can distort the underlying physical premise and lead to incorrect reasoning. To systematically study this challenge, we introduce \textbf{PhysAlign}, a benchmark designed to assess whether multimodal models correctly associate information recognized from physics diagrams with its intended physical role. By disentangling visual recognition from physical-role assignment through localized probes and controlled variants, PhysAlign isolates correspondence errors from recognition failures. It contains 3,341 human-validated probes spanning 986 physics problems, enabling systematic evaluation of visual recognition and physical-role correspondence at scale. We further introduce five complementary evaluation metrics, including CAcc, GAcc, and JAcc, which provide a comprehensive assessment of models' ability to recognize diagram content, establish correct physical correspondences, and solve the underlying physics problem. Across our evaluated multimodal models, PhysAlign reveals a consistent gap between local visual recognition and physical-role grounding. Even when the queried content is correctly recognized, the conditional correspondence error rate remains 13.8\% for GPT-6-Astra and rises to about 50.6\% for InternVL3.5-8B. These findings indicate that strong perception alone does not ensure reliable physical interpretation, exposing a distinct grounding bottleneck that is largely hidden by answer-level accuracy and highlighting the need for future models to better align recognized visual evidence with its physical meaning.
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2609.33319 [cs.AI]
  (or arXiv:2609.33319v2 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2609.33319

arXiv-issued DOI via DataCite

Submission history

From: Haoyang Liu [view email]
[v1] Sun, 27 Sep 2026 07:34:22 UTC (3,202 KB)
[v2] Tue, 17 Nov 2026 16:13:19 UTC (1,773 KB)

来源:arXiv cs.AI · arxiv.org