How can I improve on low bitrates?

[mf]

Trellis drops coefficients -- so it's probably not good for high bitrate. It is for low.

[snowbeach]

@mf

Excerpt from my guide:
Trellis based R-D quantisation (Rate-Distortion quantisation) is more experimental and it works only with the quantization type H.263 for now. It is supposed to set DCT-coefficients to 0, if the distortion introduced by that is lower than a weighted amount of bits saved (coding 0-coefficients is efficient). It will introduce a slight drop in PSNR (Peak Signal to Noise Ratio)/absolute mathematical quality by its nature, but the filesize gain overcompensates for that (in 2 pass scenarios). At this point you might use H.263 in conjunction with Trellis R-D quantisation option in case of 2 CD rips, or just stick with MPEG or MPEG-Custom and wait for Trellis R-D quantisation to come into play later.
You say the opposite. :confused: I do not know anymore, where I got this info. But I am sure I read it somewhere in the forum... How sure are you that you are right regarding this topic? :p

BTW. With 2 CD rips, I mean high bitrate solutions of course.

[sungey]

IMHO, deciding to use or not to use bframe is not about high or low bitrate. Bframe is to decrease the first-pass file size.
The closer the final video size to the first-pass size .. the better quality it will be ... just my 2 cent ...

[Selur]

@sungey:
"The closer the final video size to the first-pass size .. the better quality it will be ... just my 2 cent ..."
This might be true if one uses a unfiltered quant 2 encode, since then it means: the archived file is next to the best possible. But in general I would say your statement is wrong.

about trellis quant:
Trellis based R-D quantisation
This feature on the debug-tab is at that location for a reason: it's more experimental. It only works with h.263 quantisation matrix.
It's supposed to set DCT-coefficients to 0 if the distortion introduced by that is lower than a weighted amount of bits saved (coding 0-coefficients is efficient.).
It'll introduce a slight drop in PSNR by it's nature, but the filesize gain overcompensates for that (in 2pass scenarios, that
by Koepi (http://forum.doom9.org/showthread.php?s=&threadid=53383)

Cu Selur

[mf]

Originally posted by snowbeach
You say the opposite. :confused: I do not know anymore, where I got this info. But I am sure I read it somewhere in the forum... How sure are you that you are right regarding this topic? :p
Setting coefficients to 0 is dropping them. It's exactly the same as I say.

[snowbeach]

I see, you are pretty sure. :D I am going to change this in my guide and say that you should use it for low bitrate solutions (below 900 kBit/s). OK?

[mf]

Originally posted by snowbeach
I see, you are pretty sure. :D I am going to change this in my guide and say that you should use it for low bitrate solutions (below 900 kBit/s). OK?
Sure, I've tested trellis with B-frames at low bitrate, and there is a PSNR increase.

The logic behind it: Dropping coeffs hurts quality, but if bitrate savings are higher, it means the codec will be able to use lower quants, and retain higher quality. This is in a low-bitrate 2-pass scenario. This can still be true for high-bitrate, but less so. There are less bitrate savings so your overall quality gain might not be as big as for low-bitrate.

[sungey]

Originally posted by Selur
@sungey:
"The closer the final video size to the first-pass size .. the better quality it will be ... just my 2 cent ..."
This might be true if one uses a unfiltered quant 2 encode, since then it means: the archived file is next to the best possible. But in general I would say your statement is wrong.


Cu Selur

What i meant was ... if your first-pass stats is 100 mb ... and u aimed to encode at 30 mb .... u wont get good result (alot of high quant frames) .... but if u somehow decrease the first pass size to 65-70 mb using bframe .. u will get better result when u encode at 30 mb (as long as the quant for bframe isnt too high) .. this hold true for 2 pass encoding ..

[Selur]

"if your first-pass stats is 100 mb ... and u aimed to encode at 30 mb .... u wont get good result (alot of high quant frames) ..."
Not too sure if that is right,.. since high quantizer could still mean good quality, it's all a question of the quant matrix one chooses :D

But I see what you ment, thx for the clearyfication :)

Cu Selur

[snowbeach]

@mf

included your statement to my guide. :D it's online.

[MrBunny]

I would just like to point out quickly that "high" and "low" bitrates really depend on the source material. While 1500 kbit/s might be overkill on a really dark low motion movie, it would probably be a medium or even low bitrate on something like a noisy TV or VHS capture (assuming unfiltered). Even from one movie to another, there can be a large difference in compressibility. Because of that, I don't believe that you can really define a single bitrate as defining high vs low bitrate solution.

Personally, I don't think that there can ever be too many bits. There's always things that can be done to raise quality when you have the bits to spare. A more "accurate" quant matrix, a sharper resize, high resolution...

Just my 2 cents.

[Soc]

I just want to clarify a minor point on an issue that a few people seem confused about (not just in this thread):

Originally posted by mf
If you have quality loss from B-frames, just do 100/100/0. That will save bitrate but keep quality.Originally posted by mf
Well, think about it. B-frames at 100% quant are just P-frames that can reference future and past frames. So even if there is nothing to reference in the future, there will be no quality loss, because then it will in essence be a P-frame.

This is not exactly correct, since offset of 100 means the B-frames will have a quant that is 1 higher. IMHO they would still help the overall quality, but if you're using a very high bitrate and want them to have the same quant as a P-frame then use 100/0/0 (a better alternative to turning them off entirely)

Sorry, I know that was a little off topic, but I'm picky about what numbers are for and little things like that tend to bug me :D

Btw, I've had good results using Trellis for low bitrate encodes, haven't tried it with high bitrate.

- Soc

[mf]

Originally posted by snowbeach
@mf

included your statement to my guide. :D it's online.
lol :D

Who said you could steal my text over which I have copyright and use it in your guide ? :rolleyes: :p Not that I care ;).

[kilg0r3]

Originally posted by Soc
(a better alternative to turning them off entirely)

Yep, also because b-frames that are not quantized higher than p-frames are larger than p-frames because they store motion vectors of preceeding and succeeding frames.

[snowbeach]

Have you read my disclaimer? I like it... :p

Soc is right. Just one example for his argumentation. Hopefully I understood him right... Encoded Star Trek Motion Picture Show to 2 CDs à 700 MB without making a comp test, I was lazy, last night. The movie lasts 2h18min. This morning I got a file that is ~900 MB (CD1 716MB and CD2 179MB). It seems that this movie must be very good compressable. So I am going to encode it one more time, but only to 1 CD, with the same settings as for 2 CDs. :D Which means I am not going to increase B-frame threshold 100, not using Trellis R-D quant, no H.263 (just leaving MPEG), leaving QPel and still payback with bias (maybe I should change only this option to proportional) as described in my guide for rips with low bitrates. So I handle this 1 CD rip like a high bitrate rip (comp test 76%), because of the very good compressibility.

[symonjfox]

(a better alternative to turning them off entirely)
Soc just confirmed the things I said before and I was named as "brainless".

A new Hard disk is arriving to me, and when I'll got it, I'll test for you.

In theory a Quant 2 rip with NO Bframes should be better that a Quant 2 rip with B 100/100/0.

Wait 1 day and you'll have the final answer.

[mf]

Originally posted by snowbeach
Have you read my disclaimer? I like it... :p
“The author intended not to use any copyrighted material for the publication or, if not possible, to indicate the copyright of the respective object.”
Why am I not listed? :D
Originally posted by symonjfox
Soc just confirmed the things I said before and I was named as "brainless".
I called the decision (or recommendation) brainless. Not you. Don't take it personally.

[Soc]

@symonjfox
You misunderstood me. I was saying that you should not turn off B-frames, merely lower the ratio and offset.

The term "brainless" may have been a little harsh, since yours is not an uncommon misconception, but in future please read more carefully before jumping to conclusions. For example, you said:
"Then I read a long thread (http://forum.doom9.org/showthread.php?s=&threadid=53720&perpage=20&pagenumber=1) talking about Bframes on high bitrate and someone says that it's better disable them on high bitrate streams."

If you had read that thread properly, you would have seen that sysKin states (referring to 3/100/0/0):
"if you're afraid that b-frames will have lower quality, use this setting :) B-frames will have higher quality than p-frames"

I suggest you search the forum a little for an explaination of how B-frames work, and what the settings do, before taking up any more space in this thread discussing them.

- Soc

[mf]

Didn't I say it was a recommendation from sysKin :D. But I was indeed wrong about the offset, I don't usually mess with that particular setting. Although you probably still would get higher (or equal) quality at that offset.

[snowbeach]

You are listed, indirectly:
Many thanks go to the users of the XviD forum at Doom9, who support the extention of this guide.
And do not say I did not mention you. :D
But maybe I am going to add a credits list. :D

BTW. I forgot, I mentioned you at the Newbie Settings thread. ;)

[symonjfox]

So, sorry to everyone.
I'm not a native english user, I read posts many times, believe me. I search a lot on this forum, but when things start to become more technical, it's much more difficult to me to understend it correctly.

For example, I always use Bframes on my encodes (but I create 1 cd files). Then I read many posts on B frames, then I read that post that confused me. Then I post those ideas on this thread and I though that brainless was used for me, not for that (I saw many users being very bad with others). Sorry again.

I sometimes go to italian doom9, but the original main forum is more advanced :o

PS: I'll do the test the same, so I'll post the final results :)

CYA

[mf]

Originally posted by symonjfox
Then I post those ideas on this thread and I though that brainless was used for me, not for that (I saw many users being very bad with others). Sorry again.
No need to be that polite, I'm not either :D. Anyway, just give the ppl that recommended you the "brainless idea" a good whack - you can't help it that you get confused, they should know what they're talking about :p. And otherwise they deserve whacks from sysKin (I get those too when posting false information :D).

[bilu]

Originally posted by mf
And otherwise they deserve whacks from sysKin (I get those too when posting false information :D).

And you seem to like it :D Just kidding eheh


Bilu