BSCV数据集
收藏资源简介:
BSCV数据集是首个用于真实世界比特流损坏视频恢复的大型基准数据集,由南洋理工大学和香港理工大学合作创建。该数据集包含超过28,000个视频片段,这些视频片段是从YouTube-VOS和DAVIS数据集中提取并详细处理的。数据集通过使用流行的H.264视频编解码器压缩视频片段,并随机移除比特流中的段来模拟解码视频中的数据包丢失错误和存储损坏错误。BSCV数据集的特点是包含多种现实世界的损坏模式,如块状伪影、颜色伪影、重复伪影、错位、纹理丢失和拖尾伪影,这些都是在多媒体通信中常见的。该数据集旨在解决视频通信和多媒体取证中的实际视频丢失问题,通过提供一个包含真实和不可预测错误模式的数据集,推动视频恢复技术的发展。
The BSCV dataset is the first large-scale benchmark dataset for real-world bitstream-corrupted video restoration, co-developed by Nanyang Technological University and The Hong Kong Polytechnic University. This dataset contains over 28,000 video clips, which are extracted and thoroughly processed from the YouTube-VOS and DAVIS datasets. The dataset simulates packet loss errors and storage corruption errors in decoded videos by compressing video clips with the popular H.264 video codec and randomly removing segments from the bitstream. The BSCV dataset features a variety of real-world corruption patterns, including blocking artifacts, color artifacts, repetition artifacts, misalignment, texture loss, and smearing artifacts, which are commonly encountered in multimedia communications. This dataset aims to address practical video loss issues in video communications and multimedia forensics, and promote the development of video restoration technologies by providing a dataset with authentic and unpredictable error patterns.

- 1Bitstream-Corrupted Video Recovery: A Novel Benchmark Dataset and Method南洋理工大学 · 2023年



