Stand-alone use of 'trellis 2'

[asarian]

Am I correct in interpreting Lord Mulder's article (http://mulder.dummwiedeutsch.de/pub/x264/) on x264, thinking that 'trellis 2' without further psy- options is not recommended, and that you best use 'trellis 1' in those cases?

[Audionut]

AFAIK, trellis is beneficial in all cases. Regardless of other setting used and or trellis 1 or 2. Just that 2 is slower.

[asarian]

AFAIK, trellis is beneficial in all cases. Regardless of other setting used and or trellis 1 or 2. Just that 2 is slower.
Thanks. I was particularly referring to this statement:

"Before the Psy optimizations were added to x264, Trellis 2 was said to have a tendency to remove fine details and improve edges. Therefore Trellis 1 often was considered the better choice. But now when using Psy RDO, it's highly recommended to use Trellis 2, although it's significant slower than Trellis 1."

So it kinda leaves me wondering now what he meant by that then, if trellis 2 is always better.

[LoRd_MuldeR]

The difference between Trellis=1 and Trellis=2 is that Trellis=2 will use Trellis more often (i.e. on all mode decisions) while Trellis=1 uses Trellis less often (i.e. only on the final encode).

Consequently if Trellis=1 gives a benefit, then Trellis=2 will give even more benefit. And if Trellis=1 did hurt, then Trellis=2 would hurt even more.

I think nowadays nobody really doubts that Trellis is beneficial. This leaves "Trellis=1 -vs- Trellis=2" as a "speed -vs- quality" option.

And of course you should always use the slower/better settings (i.e. Trellis=2), if you have the time. Also note that x264 presets "slower" and "veryslow" use Trellis=2.


The first sentence in the excerpt you quoted was written, because long ago, when x264 had no Psy-optimizations, some people claimed that Trellis (and Trellis=2 in particular) has a tendency to remove fine detail.

So I wanted to make 100% clear that this, if it ever was the case, certainly isn't the case in an up-to-date x264. Hence using Trellis=2 is perfectly "save" now.

(Needless to say the Psy RDO is enabled by default in current x264. It will be enabled unless you explicitly set "--psy-rdo 0.0" or "--no-psy" or lower "--subme" below 6)

[thewebchat]

On the contrary, I always use trellis=0 for animation. Disabling trellis allows the use of deadzones, which produce much more consistent noise/detail retention on animation. It also doubles the encoding speed.

[asarian]

The first sentence in the excerpt you quoted was written, because long ago, when x264 had no Psy-optimizations, some people claimed that Trellis (and Trellis=2 in particular) has a tendency to remove fine detail.

So I wanted to make 100% clear that this, if it ever was the case, certainly isn't the case in an up-to-date x264. So using Trellis=2 is perfectly "save" now.
Thank you very much for your explanation, LoRd_MuldeR. I understand it much better now.

(Needless to say the Psy RDO is enabled by default in current x264. It will be enabled unless you explicitly set "--psy-rdo 0.0" or "--no-psy" or lower "--subme" below 6)
Not so needless, it would seem, as I erroneously thought Psy RDO was only used when specifically adding a '--psy-rd' command. :) Hence my confusion.

Again, your reply (and other docs) are much appreciated!

[Mr VacBob]

On the contrary, I always use trellis=0 for animation. Disabling trellis allows the use of deadzones, which produce much more consistent noise/detail retention on animation. It also doubles the encoding speed.

Have you tried psy-trellis? Of course, that won't improve speed.

[Audionut]

On the contrary, I always use trellis=0 for animation. Disabling trellis allows the use of deadzones, which produce much more consistent noise/detail retention on animation. It also doubles the encoding speed.

Deadzones still affect mode decision if trellis=1.

http://forum.doom9.org/showthread.php?p=1307463#post1307463

[LoRd_MuldeR]

Not so needless, it would seem, as I erroneously thought Psy RDO was only used when specifically adding a '--psy-rd' command. :) Hence my confusion.

You may want to study "x264.exe --longhelp" some more ;)