arXiv cs.AI· Chunxi Luo, Qishuai Ding, Xingchen Peng, Hengwei Ye, Ke Zhang, Wenxin Lian, Tianyi Lu, Chunlong Yu, Zhenya Lin·· 2 天前AI 评分23
Frame Evidence：多图空间推理中参考系与视觉证据的基准测试
Frame Evidence:A Benchmark for Reference Frames and Visual Evidence in Multi-Image Spatial Reasoning
AI 导读
研究者推出 Frame Evidence 基准，包含 320 组匹配三元组和 960 道中文多选题，用于检验多图空间推理中参考系与视觉证据变化时答案是否仍正确。在 288 道留出题上，GPT three-shot 单项准确率达 50.00%，但最佳三帧组准确率仅 5/48（10.42%），由 GPT two-shot 取得。少样本提示与任务适配能提升单项准确率，却未持续改善联合正确性。
正文 · 原文
Abstract:A spatial answer depends on the reference frame specified in language and its supporting visual evidence. Item accuracy alone obscures whether correctness persists when either changes. We introduce Frame Evidence, a benchmark of 320 matched triplets and 960 Chinese multiple-choice questions from captured and rendered scenes. One family compares three grounded reference frames over fixed images. The other pairs image deletion annotated as insufficient with answer-preserving addition. Joint metrics require correct answers across all three frames and paired evidence variants. Across 20 configurations on 288 held-out questions, GPT three-shot reaches 50.00\% item accuracy, but the best three-frame group accuracy is 5/48 (10.42\%), reached by GPT two-shot. Few-shot prompting and task-specific adaptation improve item accuracy without consistently improving joint correctness. Input controls show where images support matched success, while repeated executions reveal answer instability beneath similar scores.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.19564 [cs.CV] |
| (or arXiv:2610.19564v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19564 arXiv-issued DOI via DataCite |
Submission history
From: Chunxi Luo [view email]
[v1]
Wed, 14 Oct 2026 04:22:42 UTC (4,592 KB)
[v2]
Thu, 15 Oct 2026 03:11:14 UTC (4,592 KB)
来源：arXiv cs.AI · arxiv.org