JVET-AD0181 AHG8: Update on TVD object tracking dataset [Z. Liu, L. Wang, X. Xu, S. Liu (Tencent)]

This contribution provides some updates on the TVD object tracking dataset.

In the v2 version, the missed TVD-03_03 results and the link of dataset are added.

The TVD video dataset contains three videos, i.e., TVD-01, TVD-02 and TVD-03, with different content and the corresponding tracking ground truth. In previous version, the videos are compressed in lossy mode. In this contribution, the lossless raw sequence for each video are provided.

Q: Does this impact the performance of machine analysis tasks. According to cross-checker, the impact is minor (< 3% BD rate, which may still be in the range of uncertainty in this activity)

It is agreed in principle that it is approproiate to exchange the sequences, but the new sequences should get another name (e.g., “_a”), and this new name should also appear in the CTC document.

This was discussed with WG 4 in the joint meeting. It was agreed to modify the data set.