allenai/pixmo-cap-qa

PixMo-CapQA是一个关于图像的问答对合成数据集。该数据集是通过使用Claude大型语言模型从图像的密集描述中生成问答对而创建的，且模型并未看到实际图像。PixMo-CapQA是PixMo数据集集合的一部分，并用于训练Molmo系列模型。数据格式包括图像URL、问题、答案和消息字段。图像以URL形式存储，需要单独下载。问题字段包含输入文本，答案字段包含最终目标输出文本，消息字段以消息列表格式包含相同的数据。数据集采用ODC-BY-1.0许可证，适用于研究和教育用途。

PixMo-CapQA is a synthetic dataset of question/answer pairs about images, generated by using the Claude large language model from dense captions of images without seeing the actual images. The dataset includes image URLs, questions, answers, and messages fields, with image URLs being repeatable. The question field contains input text, the answer field contains the target output text, and the messages field contains the same data in a list-of-messages format. The dataset is used to train the Molmo family of models and is part of the PixMo dataset collection.