five

From single video sequence - Observations of L1 (SAD, SATD) and L2 (SSE) vs Structural Similarity (SSIM) for Intra and Inter coded frames

收藏
Figshare2016-12-20 更新2026-04-08 收录
下载链接:
https://figshare.com/articles/dataset/From_single_video_sequence_-_Observations_of_L1_SAD_SATD_and_L2_SSE_vs_Structural_Similarity_SSIM_for_Intra_and_Inter_coded_frames/4483724/1
下载链接
链接失效反馈
官方服务:
资源简介:
Observations of L1 (Sum of Absolute Difference (SAD), Sum of Absolute Transform Difference (SATD)) and L2 (Sum of Squared Errors (SSE)) against Structural Similarity (SSIM) at sub-block (4x4) and (8x8) on H.264/AVC. This taken during Intra and Inter coding. Limited to a single video, Foreman, see paper. <br> Please note the following when dataset is unzipped: SAD and SSE is marked out on the folder name, however SATD is marked as PsuedoSSIM. The observations are recorded within file that have an extension/end with .dat (dot dat). <br> Codebook: Below are the headings and an overall description. BlockSize SSIM OneMinSSIM1k Covariance AbsMeanDiff PseudoSSIM TradDist Only the first two columns and the last column were used for the paper referenced. BlockSize – this will be 4 for 4x4, 8 for 8x8. SSIM – SSIM score as calculated using the implementation found within JM H.264/AVC. TradDist – This would refer to L1 or L2 norm of SA(T)D or SSE respectively. <br>OneMinSSIM1k, Covariance, AbsMeanDiff and PseudoSSIM refer to the development of the PsuedoSSIM score, a model to map traditional distortion metric assessment to SSIM values. This is concept is explained in: http://dx.doi.org/10.1109/BMSB.2013.6621695

本数据集包含H.264/AVC(H.264/高级视频编码)帧内编码与帧间编码过程中,针对4×4和8×8子块的L1范数(绝对误差和(Sum of Absolute Difference, SAD)、绝对变换误差和(Sum of Absolute Transform Difference, SATD))、L2范数(均方误差和(Sum of Squared Errors, SSE))与结构相似性(Structural Similarity, SSIM)之间的对比观测数据。本次实验仅使用单段测试视频Foreman,详细信息请参阅相关论文。 解压数据集时请注意以下事项:文件夹名称中会标注SAD与SSE,但SATD会以PsuedoSSIM的名称标识;所有观测数据均存储在扩展名为.dat的文件中。 数据字段说明: 以下为各数据字段的名称与整体说明:BlockSize、SSIM、OneMinSSIM1k、Covariance、AbsMeanDiff、PseudoSSIM、TradDist。本次引用的论文仅使用了前两列与最后一列数据。 - BlockSize:取值为4时对应4×4子块,取值为8时对应8×8子块。 - SSIM:采用JM H.264/AVC参考实现所计算得到的结构相似性分值。 - TradDist:分别对应SA(T)D(即SAD或SATD)与SSE的L1或L2范数。 OneMinSSIM1k、Covariance、AbsMeanDiff与PseudoSSIM均与PsuedoSSIM分值的推导相关,该模型用于将传统失真度量评估结果映射至SSIM分值。相关概念的详细说明可参阅:http://dx.doi.org/10.1109/BMSB.2013.6621695
创建时间:
2016-12-20
二维码
社区交流群
二维码
科研交流群
商业服务