ReCRVD
收藏资源简介:
ReCRVD是由天津大学电气与信息工程学院创建的大规模视频去噪数据集,包含120组噪声-干净视频对,ISO值范围从1600到25600。该数据集通过重新捕获4K屏幕上显示的高分辨率视频内容,并采用高低ISO设置来构建噪声-干净配对帧,随后通过强度、空间和颜色校正策略进行对齐。数据集创建过程中,通过保持时间顺序将校准后的帧连接成配对视频,以模拟真实动态场景。ReCRVD数据集旨在为监督学习提供丰富的真实运动数据,以提升视频去噪模型的性能和泛化能力。
ReCRVD is a large-scale video denoising dataset developed by the School of Electrical and Information Engineering, Tianjin University. It contains 120 pairs of noisy-clean videos, with ISO values ranging from 1600 to 25600. This dataset is constructed by re-capturing high-resolution video content displayed on a 4K screen using both high and low ISO settings to generate noisy-clean paired frames, which are then aligned via intensity, spatial, and color correction strategies. During the dataset creation process, the calibrated frames are concatenated into paired videos while preserving temporal order to simulate realistic dynamic scenes. The ReCRVD dataset aims to provide abundant real-world motion data for supervised learning, so as to enhance the performance and generalization capability of video denoising models.




