DiffIQA
收藏资源简介:
DiffIQA是一个由香港理工大学、城市大学和OPPO研究院创建的大规模图像质量评估数据集。该数据集包含了大约180,000张图像,这些图像是通过调整扩散增强方法的超参数生成的,质量相对于参考图像有更差、相似或更好的变化。数据集中的图像均由人类进行质量标注,用于训练和评估图像质量评估模型。DiffIQA的构建旨在突破传统全参考图像质量评估中对完美参考图像质量的依赖,为图像质量评估研究提供了新的数据资源和挑战。
DiffIQA is a large-scale image quality assessment (IQA) dataset developed by The Hong Kong Polytechnic University, City University of Hong Kong, and OPPO Research Institute. This dataset comprises approximately 180,000 images generated by tuning the hyperparameters of diffusion enhancement methods, with their perceptual quality ranging from worse, similar, to better compared to the reference images. All images in the dataset have been manually annotated with quality scores, and are utilized for training and evaluating image quality assessment models. The construction of DiffIQA aims to break the limitation of traditional full-reference IQA methods that rely on perfect reference images, thereby providing novel data resources and research challenges for the field of image quality assessment.




