arXiv · cs.AI · Haseeb Yaqoob, Fatima Azfar, Emaan Arif Khan2026-11-03 13:00 · 8 小时前模型评分33

六套系统在 Urdu Nastaliq 手写识别基准上的零样本评测

Benchmarking Vision-Language Models for Urdu Nastaliq Handwriting Recognition

导读

研究者在 Urdu Nastaliq 手写数据集（UNHD）的 2,465 张单行图像上零样本评测了六套系统，包括 Tesseract、印刷体识别器 UTRNet-Large、两个开源视觉语言模型（VLM，Vision-Language Model）和两个商用 VLM。

分类置信度

中 · 待人工复核

来源级别：媒体与个人

事件类型：研究发布

证据摘录

“We benchmark six systems on the Urdu Nastaliq Handwritten Dataset (UNHD): Tesseract (default configuration), a script-specific recognizer trained on printed Urdu (UTRNet-Large), two open-source vision-language models (VLMs), and two commercial VLMs.”
“All six systems are evaluated without fine-tuning on 2,465 single-line images (Part A), and the four VLMs receive the same fixed prompt (decoding uses temperature 0 where configurable, except for Claude Sonnet 5, whose decoding cannot be configured).”
“We do not compare against handwriting-trained recognizers, which would likely do better in-domain; we characterize zero-shot behavior, not state of the art.”

来源：arXiv · cs.AI · arxiv.org