Error-Concealed Video Dataset (ECVD)

This is a dataset of 120 error-concealed video clips. The clips were generated from 6 CIF, 6 HD and 6 Full-HD test video sequences. Each of those sequences was error concealed with 4 Error Concealment (EC) techniques: Motion Copy, Motion Vector Extrapolation, Decoder Motion Vector Estimation (DMVE) + Boundary Matching Algorithm (BMA), and Adaptive Error Concealment Order Determination (AECOD). The dataset also includes the original (loss free) video clips, as well as the subjective ranking of the error-concealed videos. The original purpose for generating this dataset is to evaluate the performance of various Image/Video Quality Assessment (I/VQA) methods in how well they compare the quality of error-concealed videos. In other words, if the output of EC technique A is a better-quality video than EC technique B, which I/VQA metric predicts this correctly. For more information please refer to the following paper:M. Kazemi, M. Ghanbari, and S. Shirmohammadi, “The Performance of Quality Metrics in Assessing Error-Concealed Video Quality,” IEEE Transactions on Image Processing, accepted March 14 2020, to appear..本数据集包含120段经错误隐藏（Error Concealment, EC）处理的视频片段。这些片段源自6个CIF、6个HD以及6个全高清（Full-HD）测试视频序列。上述每个序列均使用4种错误隐藏技术完成处理：运动复制（Motion Copy）、运动向量外推（Motion Vector Extrapolation）、解码器运动向量估计（Decoder Motion Vector Estimation, DMVE）+边界匹配算法（Boundary Matching Algorithm, BMA），以及自适应错误隐藏顺序确定（Adaptive Error Concealment Order Determination, AECOD）。该数据集同时收录了原始（无损）视频片段，以及经错误隐藏处理的视频的主观评分排序。构建本数据集的初衷，是评估各类图像/视频质量评价（Image/Video Quality Assessment, I/VQA）方法的性能，具体而言，即验证这些方法能否准确比对经错误隐藏处理的视频质量：若某错误隐藏技术A生成的视频质量优于技术B，对应的I/VQA指标能否正确预测这一结果。如需了解更多细节，请参阅以下论文：M. Kazemi、M. Ghanbari与S. Shirmohammadi, "The Performance of Quality Metrics in Assessing Error-Concealed Video Quality,", IEEE Transactions on Image Processing, 2020年3月14日录用，即将刊发。

搜集汇总

背景与挑战

背景概述

该数据集包含120个经过错误隐藏处理的视频片段，源自多种分辨率的测试序列，并应用了四种不同的错误隐藏技术。其核心目的是评估图像和视频质量评估方法在比较不同错误隐藏技术输出质量时的性能，同时提供了原始视频和主观排名作为参考。