SupER数据库
收藏资源简介:
SupER数据库是由弗里德里希-亚历山大大学埃尔朗根-纽伦堡分校创建的,包含超过80,000张图像,涵盖14个场景,旨在解决模拟到真实差距的问题,特别是在超分辨率(SR)性能评估中。该数据集通过硬件合并获取,覆盖了CMOS传感器噪声、真实采样在四个分辨率级别、九种场景运动类型、两种光度条件以及五级有损视频编码。SupER数据库为量化评估提供了像素级地面实况,是现有基准在质量和数量上的一次飞跃,适用于微调基于学习的方法,并促进未来对真实图像的评估。
The SupER database was created by Friedrich-Alexander University Erlangen-Nuremberg. It contains over 80,000 images across 14 scenarios, and is intended to mitigate the simulation-to-real gap, particularly for performance evaluation of super-resolution (SR) tasks. This dataset is acquired through hardware-based integration, covering CMOS sensor noise, real-world sampling across four resolution levels, nine types of scene motion, two photometric conditions, and five levels of lossy video coding. The SupER database provides pixel-level ground truth for quantitative evaluation, representing a qualitative and quantitative leap over existing benchmarks. It is applicable for fine-tuning learning-based methods and facilitates future evaluations of real-world images.




