pixmo-point-explanations
收藏资源简介:
PixMo-Point-Explanations数据集包含图像、问题和答案,答案中可能包含指向图像部分的点。该数据集用于训练视觉语言模型,使其能够通过文本和点的混合形式回答问题。数据集是PixMo数据集集合的一部分,并用于训练Molmo系列模型。数据集被认为是实验性的,因为生成此类输出的模型可能会产生更多的幻觉。数据集的图像以URL形式存储,并包含用于验证图像完整性的哈希值。数据集的许可证为ODC-BY-1.0,适用于研究和教育用途。
The PixMo-Point-Explanations dataset contains images, questions, and answers, where the answers may include points referring to specific parts of the images. This dataset is designed to train visual-language models to answer questions via a hybrid format combining text and points. As part of the PixMo dataset collection, it is utilized for training the Molmo series of models. The dataset is considered experimental, as models generating such outputs may produce more hallucinations. Images stored in the dataset as URLs are accompanied by hashes for verifying their integrity. The dataset is licensed under ODC-BY-1.0 and is applicable for research and educational purposes.




