KVG-Bench

该数据集是一个涵盖10个领域、包含1300个精心挑选的测试用例的综合数据集，专为知识密集型视觉定位任务而设计。该数据集被用于评估DeepPerception模型的表现，并显著提升了准确性和跨领域泛化能力。具体规模上，该数据集包含了1300个测试用例，针对的是知识密集型视觉定位（Kvg）任务。

This dataset is a comprehensive benchmark tailored for knowledge-intensive visual grounding (KVG) tasks, encompassing 10 domains and comprising 1300 carefully curated test cases. It is utilized to evaluate the performance of the DeepPerception model, and has notably enhanced the model's accuracy and cross-domain generalization capabilities. Specifically, this dataset contains 1300 test cases targeting the aforementioned knowledge-intensive visual grounding task.