TIGER-Lab/ABC-VG-Instruct

ABC-VG-Instruct是一个指令微调数据集，它包含来自视觉基因组数据集的图像的指令-标题对。每个数据集元素包含4个与图像中不同边界框对应的指令-标题对，用于训练嵌入模型，以便基于指令嵌入场景的特定方面。

ABC-VG-Instruct is an instruction finetuning dataset containing instruction-caption pairs for images in the Visual Genome dataset. Each element in the dataset consists of 4 instruction-caption pairs corresponding to different bounding boxes in the image, used for training embedding models to embed specific aspects of a scene based on instructions.