Avoiding VHQ encoding noisy clips ?

[Poutnik]

Some time ago I did some compression tests with different quality clips,
quantizers, VHQ and bVHQ ON/OFF.

It seems that for noisy clip (too bad analog TV channel, difficult to filter, too good topic not to encode )
or low Q (e.g. Q2) or even both cases is better to avoid VHQ(or even bVHQ), because it produces even larger clip. (higher VHQ is worse).

well filtered ones and higher Q is the opposite case VHQ (and bVHQ) is better.
Suppose - not tested - low bitrate matrice would be affected less than high ones.

I imagine myself for former: increasing motion data size is bigger than decreasing residual non motion noisy data.
For the latter: increasing motion data size is lower than decreasing residual non motion filtered data.

Am I right ?

[sysKin]

VHQ will not necessarly give you a smaller clip, even if it usually does. It doesn't aim to decrease filesize at all (the very first implementation did that, but not anymore).

With noisy clips, all kinds of magic are possible, so it's probably best to check what looks better.

[Didée]

Generally, with increasing noise I'd decrease VHQ ... but not disable it.

It's less about filesize, more about coding efficiency, and mostly about visuals. The "coding efficiency" forced through VHQ isn't always visually appealing when the source is more noisy, because ME will find "more efficient" block coding through motion vectors that "should not be there". (In more common words: "floating wall" artefacting.)

Then, VHQ for BFrames is again a matter of taste and testing. But PFrame VHQ, I'd set to 1 for very grainy sources, or 2 for "medium" noisy ones. Completely disabling VHQ, I would not.

Sometimes it's even worth to try "silly" combinations, like ME=4 + VQH=2, or so ...

[Poutnik]

Thank you both for answers. They are more interesting and surprising than expected. I feel honoured to be answered even by two of high ranked forum members. :p