Pento-DIARef

Pento-DIARef是一个用于视觉领域中指称表达生成的诊断数据集，由波茨坦大学计算语言学系创建。该数据集包含148,000条数据，通过一个著名的符号算法——增量算法生成指称表达。数据集旨在通过简单的任务定义，即从视觉输入生成表达，测试神经模型是否能捕捉到潜在的规律并展现相应能力。数据集的应用领域在于解决指称表达生成的问题，特别是在需要唯一识别描述的场景中。

Pento-DIARef is a diagnostic dataset for referring expression generation in the visual domain, developed by the Department of Computational Linguistics at the University of Potsdam. It comprises 148,000 instances, with referring expressions generated through a well-known symbolic algorithm—the Incremental Algorithm. The core goal of this dataset is to test whether neural models can capture underlying regularities and demonstrate corresponding capabilities via a straightforward task definition: generating expressions from visual inputs. This dataset targets solving referring expression generation problems, especially in scenarios that demand uniquely identifying descriptions.