Img-Diff

该数据集旨在通过分析相似图像之间的对象差异，提升多模态大型语言模型在细粒度图像识别方面的能力。它专注于“对象替换”样本，并用于对最先进的混合多模态语言模型进行微调，已在多个基准测试中显示出性能的显著提升。虽然该数据集规模相对较小，但质量极高，其任务涵盖了图像差异识别和视觉问答。

This dataset is designed to enhance the fine-grained image recognition capabilities of multimodal large language models by analyzing object differences between similar images. It focuses on "object replacement" samples and is used for fine-tuning state-of-the-art hybrid multimodal language models, with significant performance improvements observed across multiple benchmark tests. Although the dataset is relatively small in scale, it features extremely high quality, and its tasks cover both image difference recognition and visual question answering (VQA).