MAT SDXL-Inpainting
收藏资源简介:
MAT SDXL-Inpainting 数据集由华中科技大学和维沃人工智能实验室共同构建,包含1400万张图像-掩码对。该数据集旨在解决图像修复中的复杂结构和语义一致性问题。数据集包含了丰富的前景和背景信息,并采用了Auto-Labeling框架进行标注,以确保数据质量。该数据集可用于训练图像修复模型,并已在多个公开数据集上进行评估,如Places2、CelebA-HQ和FFHQ,以验证其在结构一致性和语义一致性方面的性能。
The MAT SDXL-Inpainting dataset was co-developed by Huazhong University of Science and Technology and Vivo AI Lab, containing 14 million image-mask pairs. This dataset aims to address the challenges of complex structure and semantic consistency in image inpainting. It features rich foreground and background information, and adopts an Auto-Labeling framework for annotation to ensure high data quality. This dataset can be used for training image inpainting models, and has been evaluated on multiple public datasets including Places2, CelebA-HQ and FFHQ to verify its performance in terms of structural consistency and semantic consistency.




