OKEDIT
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/donglgcn/BalancEdit/tree/MMOKVQA
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为OKEDIT，是基于OKVQA数据集构建的，旨在评估多模态模型编辑中泛化性与局部性之间的权衡。它包含了5046个测试样本，覆盖了超过20个独特的类别。OKEDIT通过提升重述图像的质量和调整局部性样本的难度，更好地评估了泛化性与局部性之间的平衡。该数据集的规模为5046个测试样本，其任务是评估多模态模型编辑的效果。
The dataset named OKEDIT is constructed based on the OKVQA dataset, aiming to evaluate the trade-off between generalization and locality in multimodal model editing. It comprises 5,046 test samples covering over 20 unique categories. OKEDIT better assesses the balance between generalization and locality by enhancing the quality of paraphrased images and adjusting the difficulty of locality-specific samples. The core task of this dataset is to evaluate the performance of multimodal model editing.
提供机构：
Authors of the paper
