Video compression data set TVD

Learning based video compression and video analysis tasks are booming in academia and industry , High quality training and test data are urgently needed for relevant research and standard setting , For example, for JPEG AI、JVET NNVC、MPEG VCM etc. . therefore , Tencent multimedia lab has built TVD（Tencent Video Dataset） Data sets .

TVD It can be used for many tasks , For example, video coding based on neural network 、 object detection 、 Target tracking, etc .

TVD contain 86 Video sequence , It covers different contents . The resolution of each sequence is 3840x2160, Yes 65 frame . These video sequences have been used for JVET NNVC Training on relevant proposals .

For target detection tasks ,TVD Provides 166 individual 1920x1080 Image ,RGB24 Format , And calibrated bounding box. At the same time, the task is divided for the instance , And for this 166 Images provide segmentation masks annotations.

For target tracking tasks , Provides 3 Videos and corresponding label, These videos have been included in MPEG VCM Our tests focus on .

The following is a thumbnail of some sequences ,

TVD The download link for is ：https://multimedia.tencent.com/resources/tvd

You can also reply in the background “TVD” Get download link

References
X. Xu, S. Liu and Z. Li, "A Video Dataset for Learning-based Visual Data Compression and Analysis", 2021 International Conference on Visual Communications and Image Processing (VCIP)
X. Xu, S. Liu and Z. Li, "Tencent Video Dataset (TVD): A Video Dataset for Learning-based Visual Data Compression and Analysis", arXiv:2105.05961, May 2021.
X. Xu, et al. "A video dataset for training in neural network based video coding", JVET-U0116, ITU-T SG 16 Q6 and ISO/IEC JTC 1/SC 29 document, Jan. 2021. 

Interested parties, please pay attention to WeChat official account Video Coding

​