EDFace-Celeb-1M (BFR128) 和 EDFace-Celeb-150K (BFR512)

本文介绍了两个用于盲人脸修复的基准数据集：EDFace-Celeb-1M (BFR128) 和 EDFace-Celeb-150K (BFR512)。这两个数据集由北京理工大学计算机科学与技术学院的研究团队创建，包含超过150万张分辨率为128x128和512x512的人脸图像。数据集中的图像涵盖了多种降质情况，如模糊、噪声、低分辨率、JPEG压缩等，旨在为盲人脸修复算法提供全面的训练和测试资源。此外，数据集还固定了训练和测试样本的划分，确保了研究的可靠性和重复性。这些数据集的应用领域包括人脸识别、检测等视觉任务，旨在解决真实世界中人脸图像降质问题。

This paper introduces two benchmark datasets for blind face restoration: EDFace-Celeb-1M (BFR128) and EDFace-Celeb-150K (BFR512). These two datasets were developed by the research team from the School of Computer Science and Technology, Beijing Institute of Technology, and collectively contain over 1.5 million face images with resolutions of 128×128 and 512×512. The images in the datasets cover various degradation scenarios including blur, noise, low resolution, JPEG compression and others, aiming to provide comprehensive training and testing resources for blind face restoration algorithms. In addition, the datasets have fixed the training and test sample splits to ensure the reliability and reproducibility of related research. The application fields of these datasets cover visual tasks such as face recognition and detection, with the goal of solving the degradation problems of face images in real-world scenarios.