q2 encode with slight filtering or 2-pass encode with no filtering?

[Chainmax]

I am about to make a 2CD rip from one of my DVDs. When testing the script to be used on the encode (just IVTC and cropping+resizing), I saw that a constant q2 encode with b-vops set at 1/1/1 and the HVS_Best matrix yielded an average bitrate of about 1450kbps. Now, in order to make the encode fit in two discs I need an average bitrate of about 1200kbps. What do you guys think would be a better idea: slight filtering [something like RemoveGrain(mode=2).Deen("a2d"1,3,5)] and constant q2 or two-pass encode with no extra filtering and a lower bitrate matrix like H.263? The source is a pretty dark cartoon, if that helps.

[Teegedeck]

Hi Chainmax,

please excuse that I mainly want to comment on your XviD-settings. Well, those are important for the quality, too.

1) 1/1/1 for a b-vop setting is senseless and quality-degrading. This is no subjective statement, it is proven fact. Use the defaults.
2) Using XviD 1.1 with VHQ for b-frames would of course be very advantageous.
3) HVS matrices are not really a good choice for cartoons. Use H.263 or the CG matrix.
4) Don't denoise if there isn't excessive noise (that's my view). IF you want your encode to look like the original. I am aware that most anime encoders have an aesteticism of their own that is... peculiar.


Good luck!
Tee

[Chainmax]

I always use the defaults, but since b-vops at 1/1/1 seem to be used in high resolution encodes like Soulhunter's Revolutions one I figured it was a good choice for maximum quality. I haven't tried Xvid 1.1 because Koepi hasn't released a build of it yet. Is there any place as reliable as koepi's site where it can be downloaded from?

In any case, I doubt that just switching to H.263 and using b-vops at default will lower the bitrate by 250kbps, so what do you think is better: constant q2 and filter slightly or 2-pass encode with no filtering?


P.S: is the FOX matrix any good for HQ encodes of animated material?

[Teegedeck]

B-frames at offset=1 are quite OK for high-quality encodes like Soulhunter's where filesize isn't so important (though it still is a waste). But you seem to need every bit you can save.

Try for example gamr's site for 1.1 builds - or celticdruid's.

If you aim for a fixed filesize, using 2-pass is more efficient than a mixture of blind guessing and filtering.

Filtering does far more harm to an encode than the occasional high quantizer.

I haven't tried FoxHome on anime.

[Chainmax]

I just read the "vhq for beframes" thread and apparently celticdruid's builds have an up-to-date xvid.ax file, so I will try them. Thanks for all the tips :).

[Soulhunter]

Originally posted by Teegedeck

1) 1/1/1 for a b-vop setting is senseless and quality-degrading. This is no subjective statement, it is proven fact.

Uhm...

Excuse me, but where was this proven ???


I know some PSNR tests -> B-VOP -vs- P-VOP @ Q3, but...

But my own experience is quite different from this results !!!


While...

Q2-I/Q2-P/Q2-B seems to be a worse than Q2-I/Q2-P/No-B

Q2-I/Q2-P/Q3-B is better than Q2-I/Q3-P/No-B (visually) !!!


Sorry, but thats my opinion... :o


Btw, for "clean" anime I would use H.263 as well !!!

If its really noisy, I would use Deen for cleaning... ;)


My http://img49.exs.cx/img49/2229/twocents.gif - Bye

[Teegedeck]

Got me. :D I was pretty confident that you would jump at it. B-frames at offset=0 is really proven to be worse than no b-frames at all. That was the actual setting which we had discussed in-depth in one thread. B-frames at an offset of 'just'1 are still worse than the default settings but I should have given a more elaborate answer.

1 consecutive b-frame which is just compressed at p-frame-quant+1 gives you hardly any advantage in compression (at least that's what my initial tests with b-frames showed me) when compared to using only p-frames and may look slightly worse than a p-frame only encode if you don't use VHQ for b-frames. No good trade-off IMHO. I'd either use the default ratio or no b-frames at all.

[Soulhunter]

Maybe our different bitrate-aims are the culprit again ???

Hehe, another comparison to do... :D


Bye

[Chainmax]

Teegedeck, what about 6of9_HVS? Have you tried it on animated material?

[Dali Lama]

Originally posted by Chainmax

P.S: is the FOX matrix any good for HQ encodes of animated material?

I have done testing with it and I would say yes if the bitrate is high enough. It doesn't need an extreme bitrate; perhaps 2000kbs. Although, I would agree with most anime/cartoon encoders and choose h.264 for high quality to compression ratio.

I would just like to add that the FOX matrix does handle noise well, but not as well as h.264.

Sincerely,

Dali Lama

[Soulhunter]

Guess Dali Lama means H.263...

Imo its strength are "flat" surfaces, and this you have a lot in anime !!!


Bye

[SpikeSpiegel]

Originally posted by Teegedeck
1 consecutive b-frame which is just compressed at p-frame-quant+1 gives you hardly any advantage in compression (at least that's what my initial tests with b-frames showed me) when compared to using only p-frames and may look slightly worse than a p-frame only encode if you don't use VHQ for b-frames. No good trade-off IMHO. I'd either use the default ratio or no b-frames at all.

I think that it's impossible to find the ideal b-vops settings for all cases;
this is because the quantizer scale is not linear: for example, as you say, a +1 quant. difference may be not enough...but only if average quant is 3, 4, or higher.
Using high quant., as Soulhunter does, a +1 difference is certainly more aggressive (for q1-q2 diff. is not the same of q3-q4 one)

By the way, I noticed that a big quality difference between Bframes and Pframes creates a really annoying visual effect (like flickering or smoothing) on low-motion zones.
Again, the compressibility gain of the default settings is undeniable.

[Teegedeck]

Originally posted by SpikeSpiegel
By the way, I noticed that a big quality difference between Bframes and Pframes creates a really annoying visual effect (like flickering or smoothing) on low-motion zones.
I hope you haven't found that on XviD 1.1?

[SpikeSpiegel]

Originally posted by Teegedeck
I hope you haven't found that on XviD 1.1?
Nope, I'm still using v.1.0.2

...and, before you ask it, I used VHQ=4 for tests: probably, using VHQ for b-frames, this problem could disappear ;)

I wanted to test it but xvid.gamrdev.com is down...

[Chainmax]

Try celtic_druid's builds, I recently downloaded the latest version from his site.