2pass crf possibility?

[fakey]

I got nitpicking and possibly annoying question for x264.

Now that we have --direct auto can we have 2pass crf also?

2pass and not specifying bitrate already produces something like 2pass --crf 26 encoding, but hey more flexibility is always welcomed.

Anybody interested?

I'd like it when I encode some Music Videos;)

edit:typo

[akupenguin]

2pass + cqp is exactly the same as 1pass + cqp, and if you don't specify bitrate then the default is cqp=26. (Yes, that's a feature. I abuse it in my tuning. No, you shouldn't ever do that in a real encode.)

crf inherently doesn't work in 2pass. I could, however, add an "average qp" mode. This would not have crf's feature that crf(high motion) + crf(low motion) == crf(high motion + low motion), but on the scale of a whole movie it might be good enough.

[MeteorRain]

i think fakey is talking about using 2pass to optimize b-frame direct method. spatial or tempral method are determined in 2nd-pass.

[fakey]

Yes, I am.

@aku

Sorry I abused x264:D .

I did a tiny(~100frames) test the other day when --direct auto was implemented in x264 and it gave me output different from original crf encoding.

Will return with more (hopefully precise) results.

[berrinam]

What you could try is doing a first pass with crf set to output stats, and then do a second abr pass with that bitrate and that stats file. foxyshadis requested that a while ago in MeGUI, but we never got around to it.

[fakey]

Okay, seems like something stupid was stuck in my skull to make me confuse cqp result and crf one.:stupid:

Average QP? 2pass average target QP?
That would be the sweet thing that I want.

What you could try is doing a first pass with crf set to output stats, and then do a second abr pass with that bitrate and that stats file.
1crf+2pass right?
1st pass to get bitrate, 2nd to get stats, and final.
Hmm. Long way to go but for now that seems to do the trick.

[berrinam]

1crf+2pass right?
1st pass to get bitrate, 2nd to get stats, and final.
Hmm. Long way to go but for now that seems to do the trick.
Simpler:
1st pass: get bitrate and stats
2nd pass: use the slightly more advanced 2pass algorithm.

[fakey]

Oh, I thought crf doesn't output stats.
(as it doesn't accept one, that was just from my convention)

Thank you for clarifying.

[foxyshadis]

You can still specify --stats with any output mode, MeGUI just doesn't have any special support for it. (If you do use it as a one-pass size predictor, don't bother with a hardcore encode, just make it fast and a couple steps above the actual crf you would encode at to make up for the bitrate boost of a fast encode.)