CRVD 视频数据集

CRVD 全称 Captured Raw Video Denoising，是一个动态视频数据集。数据集包含 55 组，ISO 值从 1,600 到 25,600 不等的噪音干净的视频。通过研究视频帧的时间、空间和信道相关性，原始视频去噪网络 (RViDeNet) 被建立起来。原始视频被打包成四个子序列，即 RGBG 序列，分别被 RViDeNet 单独去噪，并最终融合成一个干净的视频。该数据集不仅输出原始去噪结果，还通过图像信号处理（ISP）模块输出 sRGB 结果。该视频去噪方法优于目前最先进的视频和图像去噪算法。CRVD, whose full name is Captured Raw Video Denoising, is a dynamic video dataset. It contains 55 sets of raw videos with noise levels corresponding to ISO values ranging from 1,600 to 25,600. By investigating the temporal, spatial and channel correlations of video frames, a raw video denoising network (RViDeNet) was proposed. The raw videos are split into four sub-sequences, namely RGBG sequences, which are individually denoised by RViDeNet, and finally fused into a clean video. This dataset not only outputs the raw denoised results, but also generates sRGB results via an image signal processing (ISP) module. This video denoising method outperforms the current state-of-the-art video and image denoising algorithms.

搜集汇总

背景与挑战

背景概述

CRVD（Captured Raw Video Denoising）视频数据集是一个专注于原始视频去噪的动态视频集合，包含55组ISO值从1,600到25,600不等的噪音和干净视频，总大小为20.5 GB。该数据集用于支持原始视频去噪网络（RViDeNet）的研究，通过时间、空间和信道相关性分析，输出去噪结果和sRGB格式，适用于深度学习和视频处理领域的算法开发与评估。