Bug on signalling of some partitioning layouts in VTM-9.0 encoder

[vdrugeon]

The VTM encoder initialises the value of the variable numExpSliceInTile in function HLSWriter::codePPS to pcPPS->getNumSlicesInTile(i) - 1. This variable is then used to determine the syntax element num_exp_slices_in_tile[i] written in the bitstream.

But initialising this variable to pcPPS->getNumSlicesInTile(i) - 1 can lead to issues when a tile contains several slices for which the last slice is bigger than the previous one: for example, if a tile has a height of 5 CTUs and contains two slices with the first slice having a height of 2 CTUs and the second a height of 3 CTUs. In that case, the current VTM encoder will set the value of num_exp_slices_in_tile to 1 instead of 2, which means that the decoder will derive a tile having 3 slices instead of 2 with the first two slices having a height of 2 CTUs and the last one a height of only 1 CTU.

Initialising the variable numExpSliceInTile as follows would solve the problem for the above described configuration:
uint32_t numExpSliceInTile = (pcPPS->getNumSlicesInTile(i) == 1) ? 0 : pcPPS->getNumSlicesInTile(i);

[vdrugeon]

Corresponding merge request:
​https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/-/merge_requests/1657

[biaowang]

I think the title is somehow misleading.
According to your description, it is a bug of tile/slice implementation, but not subpicture.
If subpicture is disable, is the bug still valid?

[vdrugeon]

The bug is probably still valid if subpictures are disabled, but I have only tested it with subpictures.
Can you confirm that you are happy with the bug fix or otherwise propose another fix?

[biaowang]

could you share the configuration file?
I can test for the case when subpicture is disabled.
If the bug is still valid, I request to change the name of the title, since it is not true.

[vdrugeon]

I don't have a strong opinion about the title of the bug to be honest, so I have changed it since it seems to be so important to you...
Configuration file will follow

[biaowang]

Thanks for sharing the configuration file.
Just tested and the bug is still valid when subpicture is disabled, and your MR fix the issue.

Don't get me wrong.
If a bug occurs to slice/tile implementation but subpicture is complained, it sends wrong message to the community.
IMHO, we shall keep the bug track description as precise as possible, as we do for the meeting note.

[ksuehring]

The MR is merged