DREAMBENCH++
收藏资源简介:
DREAMBENCH++是由清华大学等机构合作创建的一个大规模个性化图像生成数据集。该数据集包含150张图像和1350个提示,旨在通过多样化的图像和提示来评估个性化图像生成模型的性能。数据集的创建过程涉及从网络收集图像,并通过模型和人工评估进行筛选,以确保图像质量。DREAMBENCH++的应用领域主要集中在个性化图像生成的评估和改进,特别是在解决图像与文本提示的一致性和概念保留的问题上。
DREAMBENCH++ is a large-scale personalized image generation dataset co-created by Tsinghua University and other institutions. It includes 150 images and 1350 prompts, aiming to evaluate the performance of personalized image generation models via diverse images and corresponding prompts. The dataset's creation process involves collecting images from the web, and filtering them through both model-based and human evaluations to ensure image quality. The main application fields of DREAMBENCH++ focus on the evaluation and improvement of personalized image generation models, particularly in addressing issues related to the alignment between generated images and text prompts as well as concept preservation.




