JVET-AD0186 AHG8: VVC tool analysis for Machines Vision Tasks [Z. Liu, L. Wang, X. Xu, S. Liu (Tencent)]

This contribution analyses of the new tools in VVC in terms of the impact on the coding efficiency for machine tasks and the encoding/decoding complexity. Based on the experimental results, it is shown that the VTM can be configured with 70% complexity off while maintaining the detection accuracy on the object detection task.

In the v2 version, the results are updated.

This included interesting information with very detailed per-tool analysis.

It was commented that it is interesting to see that with some tools switched off, the machine analysis performance is increasing. However, some differences are small, and in some cases tools are beneficial for for object detection, ans not beneficial for object tracking. ALF is one example that is not beneficial for both (at least in RA). Further study was needed, particularly:

• To potentially re-define the anchor configuration
• To potentially identify which tools might be recommended to be switched off for machine analysis tasks (in the TR)

Further study in an AHG was encouraged.

AHG10: Encoding algorithm optimization (4)

Contributions in this area were discussed at 1840–2015 on Tuesday 25 April 2023 (chaired by Y. Ye).