OpenDCAI/dataflow-mm-region_caption

该数据集是通过预处理源数据集生成的，适用于图像文本对话生成、多模态大模型微调和视觉问答（VQA）等任务。数据集包含预处理后的图像文件和JSONL格式的注释文件，注释文件中每条记录与一张图像相关联，包含用户和助手之间的对话内容，描述了图像中特定区域的内容。数据集的源数据集是ReferItGame数据集，提供了高质量的自然场景图像和对象参考任务的基础材料。

This dataset is generated by preprocessing the source dataset and is suitable for tasks such as image-text dialogue generation, multimodal large model fine-tuning, and Visual Question Answering (VQA). The dataset includes preprocessed image files and a JSONL annotation file, where each entry is associated with one image and contains dialogue content between the user and the assistant, describing the content of a specific region in the image. The source dataset is the ReferItGame dataset, which provides high-quality natural scene images and foundational materials for object reference tasks.