spacepxl/personalization-meta-dataset

该数据集旨在收集用于测试图像生成模型（如Stable Diffusion）微调或个性化的数据集。这些数据集可以作为开放标准，用于比较训练设置、模型等。此外，这些数据集还可能用于实际的基准测试，如FID、面部嵌入距离、偏好评分等。目标是收集多样化的样本数据集，以反映人们实际想要训练的内容，而不是研究论文中引入的少数简单示例。

This is a collection of datasets used to test finetuning or personalization of image generation models like Stable Diffusion. These datasets aim to serve as an open standard for comparing different training settings, models, etc. Additionally, these datasets could be used for actual benchmarking, such as FID, face embedding distances, preference scores, etc. The goal is to gather a variety of sample datasets that reflect what people actually want to train, rather than just the few trivial examples introduced by research papers.