gMAD in-the-wild dataset

gMAD in-the-wild dataset是一个大规模的真实世界图像数据集，包含1,000,000对图像及其伪标签。该数据集由安徽大学互联网学院的陈元等人提出，用于训练跨数据集鲁棒的盲图像质量评估模型。数据集中的图像涵盖了多种场景和质量指标，如亮度、色彩丰富度、对比度、噪声和锐度等。创建过程中，首先从公共多媒体数据库YFCC100m中选择图像，然后通过特定的策略生成伪标签。该数据集主要应用于解决盲图像质量评估中的泛化能力和鲁棒性问题，特别是在不同数据集和真实世界场景下的图像质量评估。The gMAD in-the-wild dataset is a large-scale real-world image dataset containing 1,000,000 image pairs and their pseudo-labels. Proposed by Chen Yuan et al. from the School of Internet, Anhui University, this dataset was developed for training cross-dataset robust blind image quality assessment models. The images in the dataset cover diverse scenarios and quality metrics, including brightness, color richness, contrast, noise, sharpness, and others. During its creation, images were first selected from the public multimedia database YFCC100m, followed by the generation of pseudo-labels via specific strategies. This dataset is primarily applied to address the issues of generalization capability and robustness in blind image quality assessment, particularly for image quality assessment across different datasets and real-world scenarios.