Vereinsamt/RefCOCOg
收藏官方服务：
资源简介：
这是一个格式化版本的RefCOCOg数据集，用于大规模多模态模型评估套件（Large-scale Multi-modality Models Evaluation Suite）中，通过lmms-eval管道实现对大规模多模态模型的一键评估。数据集包含多个特征，如问题ID、图像、问题、答案、分割信息、边界框、拥挤标志和文件名，并分为验证集和测试集，分别有7573和5023个示例。它基于RefCOCOg原始数据，旨在加速多模态模型的发展。
This is a formatted version of RefCOCOg, used in the Large-scale Multi-modality Models Evaluation Suite to allow for one-click evaluations of large multi-modality models via the lmms-eval pipeline. The dataset includes features such as question_id, image, question, answer, segmentation, bbox, iscrowd, and file_name, with splits for validation and test containing 7573 and 5023 examples respectively. It is based on the original RefCOCOg data and aims to accelerate the development of multi-modality models.
提供机构：
Vereinsamt
