SridBench
收藏资源简介:
该数据集名为SridBench,是首个针对科学图表生成的基准测试,包含了来自13个自然科学和计算机科学领域的领先学术论文中的1,120个实例。这些实例在六个维度上进行评估,包括语义保真度和结构准确性。此外,该数据集还包括了在九个计算机科学方向中各选取的100个三元组,以及来自《自然》杂志最新评论和分析部分的220个三元组,确保了数据的质量、时效性和多样性。该数据集的规模为1,120个实例,其任务是科学插图生成。
This dataset, named SridBench, is the first benchmark dedicated to scientific diagram generation. It comprises 1,120 instances in total, sourced from leading academic papers across 13 disciplines of natural sciences and computer science. These instances are evaluated on six dimensions, including semantic fidelity and structural accuracy. Specifically, the 1,120 instances consist of 100 triplets selected from each of nine computer science subfields, as well as 220 triplets from the latest review and analysis sections of *Nature*, which ensures the dataset's quality, timeliness, and diversity. The core task of this benchmark is scientific illustration generation.




