MIORe & VAR-MIORe
收藏资源简介:
MIORe和VAR-MIORe数据集是针对当前运动恢复基准中的关键限制而设计的。这两个数据集通过高帧率(1000FPS)采集和专业级光学,捕捉了广泛的运动场景,包括复杂的自我相机运动、动态多主体交互和深度相关的模糊效果。MIORe通过自适应平均帧,根据计算的光流指标生成一致的动态模糊,并为视频帧插值和光流估计保留了清晰的输入。VAR-MIORe进一步扩展了范围,涵盖了从最小到极端的各种运动幅度,这是第一个明确控制运动幅度的基准。这两个数据集为下一代图像和视频恢复研究提供了挑战现有算法的高分辨率和可扩展的地面真实情况。
The MIORe and VAR-MIORe datasets are designed to address key limitations in existing motion restoration benchmarks. These two datasets capture a wide range of motion scenarios via high-frame-rate (1000 FPS) capture and professional-grade optical systems, including complex egocentric camera motions, dynamic multi-agent interactions, and depth-dependent blur effects. MIORe generates consistent dynamic blur via adaptive frame averaging based on computed optical flow metrics, while preserving sharp inputs for video frame interpolation and optical flow estimation. VAR-MIORe further expands the scope, covering a full spectrum of motion magnitudes ranging from minimal to extreme, and it represents the first benchmark that explicitly controls motion magnitude. These two datasets provide high-resolution and scalable ground truth for next-generation image and video restoration research, which challenges current state-of-the-art algorithms.




