A bunch of encoding questions

[Chengbin]

I'm curious, do you guys prefer an encoding that preserves grain from the source, or the look of a degrained video with MVDegrain 3? How about 5000Kbps 720p for both videos.

In H.264, except 8x8dct, what are the other high profile features? Do they increase quality significantly like 8x8dct? How much am I missing if I'm using main profile only?

I know this belongs to avisynth forum, but is it a good idea to use MVDenoise? Can it hurt quality in some instances? Does it remove detail (yes, but a little or a lot?).

Someone told me they use an AQ and psy-rd value of 2 because it preserves more detail, is that true?

If I use a merange 32 instead of 16, will there be improvements for most video? How much time does it kill?

What are the benefits of using 6,4 as deadzone values instead of the default 21, 11?

And finally, can you look at this video? Is this a "good" video encoded with only 520Kbps? I'm open to suggestions.

http://www.megaupload.com/?d=BC5Q8P6C

Here is the same clip, but with double the bitrate, without MVDegrain 3.

http://www.megaupload.com/?d=QBDU15H5

Tell me which one you prefer. For me, I prefer the degrained video. The grains are so clearly shown on my Archos 5's screen it is distracting (Sometimes it is bad to have too good of a screen)

[Adub]

Use search.

Oh, and asking us which ones "we prefer" doesn't matter. If you like it, then stick with it. It doesn't matter if we like it or not. You are the one watching it.

[Blue_MiSfit]

I initially replied via PM, but I'll post my reply here for posterity's sake:

1) I prefer preserving original film grain accurately about 90% of the time, unless it's a really really old movie, or the transfer was done poorly. When I degrain, I usually go down to 720p, but that's just me.

5mbps is more than enough to have good looking 720p video. In fact, most sources don't need that much. You can do surprisingly well at 5mbps for 1080p with reasonably compressible movies.

2) 8x8dct is pretty huge. It's probably the single "best" thing you can turn on with x264. So, you're missing a lot by not having it enabled. I think this is the only real part of high profile you miss out on. Not 100% sure though - I've never messed with any main profile stuff (except for QuickTime encoding, and then I just use a MeGUI preset)

3) MVDenoise is depricated MDegrain (not MVDegrain) are the standard now, per MVTools2. MDegrain is amazing, there's no substitute for it. I use this for any heavyweight denoising work that fft3dgpu can't handle effectively, or when I have lots of time to kill. Multithreaded it can run reasonably well, but it still means a good 24-30 hours for a 3 hour 1080p movie on my 3 GHz Q6600.

Any denoising will remove detail, but MDegrain is strikingly good at improving compressibility without hurting details too much - provided you tweak the thSAD values. The default of 400 is a bit high for high detail sources IMO, but is well suited to cleaning up a real mess It also pretty much never introduces banding like fft3dgpu can. When using fft3dgpu I always follow it with gradfunkmirror(1.51) to dither - which is a little wrapper script for gradfun2db that fixes the latter's border issues.

4) Always use AQ and Psy-RDO. They're both absolutely amazing. I can't stand looking at old encodes I did without these. Flat areas in particular benefit HUGELY from these features. As far as a "psy-rd value of 2", I think that refers to an old AQ mode. Not sure though. I'd leave Psy-RDO at its default of 1.0. On that note, I would also enable Psy-Trellis by turning trellis to 1, and setting the psy-rd value to 1.0:0.2 to start. The second number (0.2) is the Psy-Trellis strength. 0.2 is a good place to start, but you can increase it.

5) I've never tested ME range outside of 16, but I know its hugely slower. Not sure how much of an advantage it would provide. The only way to tell is to do a pair of 2 pass encodes, one at 16 and one at 32, at the same bitrate and compare visually.

6) Deadzones aren't necessary when using Trellis quantization. Don't bother.

7) I think your first encode is good-ish, considering it's ~500kbps. That being said, I'd never tolerate such quality In this day and age of cheap storage, why suffer such a fate?

The second one is a terrible waste of 1500kbps. You can certainly do better. Try again!

~MiSfit

[nm]

Very good and thorough reply, Blue_MiSfit!

The second one is a terrible waste of 1500kbps. You can certainly do better. Try again!
1500 kbps? The clip I downloaded was cut from a 2-pass encode made at 941 kbps, and the clip has an average bitrate of ~1200 kbps (since it contains complex scenes).

[Blue_MiSfit]

My mistake :) Still, there's room for improvement :p

~MiSfit