How can I improve on low bitrates?

[dihelson]

Hello, Friends,

Low bitrates for me: around and Below 1000...
I'd like suggestions on how can I improve the quality of my movies made on only 1 CD. Usually I'm making them on 2 CDs, which make them very good, but for 1 CD only I don't know wheather these parameters are good (I mean really good, you know! :) ), so, how can I improve my parameters for 1 CD ONLY on low bitrates, which are these:

6-Ultra High
MPEG on 2 passes (perhaps should I change this...)
Xvid
1-Mode decision
300
1
Use chroma Motion
Quarterpel
2
150
100
0
DX-50 Bvop compatible

Quantization: 2,4,2,12
Debug: Chroma Optimization ON (frame drop ratio ZERO)

Any suggestions are welcomed.

Thanks in advance,
Dihelson

[mf]

Set H.263, VHQ 4, and don't limit quants. You're about ok with the rest :).

[Vash]

You might try another quantization matrix as well. I use Hvs-good-picture for 1 CD encodes: Qualiy is nice however filesize stays small ;)

[kaitsuburi]

Leave quantizer restrictions at default (2-31-2-31). Up B-frame threshold to 100 or above.

Also, perhaps read the newbie sticky and look over the guide Snowbeach put together ;) It's excellent!

-kaitsuburi

[hellgauss]

I think the best way to encode at low birate is using avisynth filter like deen,convolution3d,ecc....

Also i suggest to choose the best resolution by doing a compressibility test at quantizer=2.

Bye:)
HG

[K_R]

Changing the quant matrix can have a dramatic effect on the size of the resultant file. Try doing a search for custom matrix. There was a thread with a zipped collection of them.

Also, maybe try bf q ratio 200...

[iago]

You might also wanna try UnFilter and UnDot -two great filters by Tom (trbarry)- to increase compressibility.

Imho comp. test (or a full 1st pass :D) is a must to decide on the parameters/filters and resolution.

For b-frames you may try 2/150/100/255 or 3/150/100/100 depending on your needs.

iago

[darksider]

Use Convolution3D filter and resize video to max 256 pixel high.
And use B frames 3, or you can try with 6 (I have good result with B frame 3 and 6).:)

[kastro68]

Koepi also recommended Undot and lumafilter and maybe convolution3d on hq setting.

BUT, I read on avisynth forum somewhere that it is bad to use undot for a clean source...because you are trying to remove random dots that aren't there. I can't remember the name of the guy that posted it, but he also went to say that UNDOT is useless because the mosquito noise is put there by the encoding process...or something like that.

i wasn't sure if it was correct or not, he made it sound very convincing.

[bilu]

Once upon a time I believed in low-bitrates for every movie with the right denoising. I was neglecting color and movement information :rolleyes:

My old learning thread, hope it is useful:

http://forum.doom9.org/showthread.php?s=&threadid=47053

Best regards,

Bilu

[cult]

@dihelsonDebug: Chroma Optimization ON (frame drop ratio ZERO)
Just a minor correction.The frame drop ratio has nothing to do with chroma optimization.I made the same mistake but mf cleared it out for me

[JohnMK]

Originally posted by K_R

Also, maybe try bf q ratio 200...

Before you do that, I'd suggest increasing the threshold to 100, from the default of 0. That will increase compressibility appreciably with barely any discernable quality loss, while a b-frame quantization ratio of 200 might be noticeable. In my own objective testing I found increasing the threshold to 255 barely increases compressibility over a threshold of 100, while the difference going from 0 to 100 is substantial. Giving deference to sysKin's belief that a threshold of 255 is risky, I think you can see why I'm recommending b-frame threshold 100.

[dihelson]

Well, Well... That's good!

Thank you all people for these nice posts.
After these many posts, I have noticed some common parameters and I think that when many people speak the same thing, they can't be all wrong, so here is a little compilation (till now) from the tips:

THE FORMULA:

======================================================
FOR 1 CD only:

6-Ultra High
CORRECTED: H.263 quant
Xvid
CORRECTED: 4
300
1
Use chroma Motion
Quarterpel
2
150
100
Threshold: CORRECTED: put 100
DX-50 Bvop compatible

Quantization: CORRECTED: Leave Default!
Debug: Chroma Optimization ON
(frame drop ratio ZERO)

Filters ? No common understanding yet. Let's try...
======================================================

I will try that on my next fim and see the new benefits and will post here the results. Thank you all people and please continue talking...

[hellgauss]

Maybe if the bitrate is very low it's better to turn off qp (it's my opinion)

HG

[dihelson]

Originally posted by hellgauss
Maybe if the bitrate is very low it's better to turn off qp (it's my opinion)

HG

Hey, this is very interesting. It would be nice if someone could comment your idea, or you could explain to us why you think so...

[thejho]

Bilu, your thread is very interesting, but in the end, how did you get the best results ( with wich filters).
I tried using fluxsmooth on the last of the mohican on a 1cd rip with ogg soundtrack, bitrate about 800, and I can't help feeling it has worse results than with none. I definitely lose detail, it's slightly blurry, and it also makes very dirty dark scenes with lots of blocks.
I'm wondering if I might not be better off with no filtering at all, except unfilter and undot that I recently discovered.

BTW, would lumafilter help for these dark scenes?

thejho

[bond]

Originally posted by dihelson
Hey, this is very interesting. It would be nice if someone could comment your idea, or you could explain to us why you think so... because qpel means a more precise motion estimation -> which leads to a sharper picture -> needs more bitrate

[snowbeach]

THE FORMULA:

======================================================
FOR 1 CD only:

6-Ultra High
CORRECTED: H.263 quant
Xvid
CORRECTED: 4
300
1
Use chroma Motion
Quarterpel
2
150
100
Threshold: CORRECTED: put 100
DX-50 Bvop compatible

Quantization: CORRECTED: Leave Default!
Debug: Chroma Optimization ON
(frame drop ratio ZERO)

Ha, ha, ha, ha, ... :D
Wait, let me think, is there a Newbie settings sticky at this forum?
Wait, let me think, you can download a guide there?
Wait, let me think, "THE FORMULA" is described in this guide?
Wait, let me think, why posting then questions like this, when everything is already there? ;)
Wait, ok - we were all new to some stuff... :D

Thumb rule: Searching the web for appropriate info:
Worst case: 1-3 hours
Best case, or should I say in this case: 15minutes ;)

Most worst case: Encoding with different settings, comparing results, etc. at least 24hours ;)

[bilu]

Originally posted by thejho
Bilu, your thread is very interesting, but in the end, how did you get the best results ( with wich filters).
I tried using fluxsmooth on the last of the mohican on a 1cd rip with ogg soundtrack, bitrate about 800, and I can't help feeling it has worse results than with none. I definitely lose detail, it's slightly blurry, and it also makes very dirty dark scenes with lots of blocks.
I'm wondering if I might not be better off with no filtering at all, except unfilter and undot that I recently discovered.

BTW, would lumafilter help for these dark scenes?


The only filtering I use is Deen("a3d",1,10,12).

Why don't you have a look at the tool I'm developing? ;)

Check my sig.

[EDIT] you should care more about compressibility and less about bitrate.

[EDIT2] Fluxsmooth recommended settings:
Fluxsmooth(3,3)
Fluxsmooth(5,5) for noisy movies/lower compressibility

[EDIT3] But lesson nr.1 is: even if you kill detail below a decent level, color and movement information will allways be important players and they can stop you from fitting in 1 CD.

Bilu

[thejho]

Thanks a lot for your prompt answer!
I took a look at your project (actually read the whole 5 pages! :p )
I'm not quite sure about how your it works (a bit OT, but what the heck !).
From what i got, you give it a min and max compressibility, resolution, a number of cd, and then it does several compressibility checks to know what settings are best to obtain the compressibilty you asked for(those settings being : resolution, denoising (?), resizing filter (?))
also thanks for the fluxsmooth settings, i used (5,5) which i gather is a bit too strong!
:rolleyes:

Good luck !

[dihelson]

Originally posted by snowbeach
Ha, ha, ha, ha, ... :D
Wait, let me think, is there a Newbie settings sticky at this forum?
Wait, let me think, you can download a guide there?
Wait, let me think, "THE FORMULA" is described in this guide?


:D :D :D
ahahahah
I prefer your "best option" 15 minutes!
Indeed, snowbeach, the "Newbie Settings" is good (is it made for 1 CD or 3 CDs ? because I think it may be different situation when you encode for 1 CD and when you encode for 3 CDs...:) ), as I said, I always made my Xvid in 2 or 3 CDs, that's not the case here, where I'd like to know explicitly some details for 1 CD ONLY, so I think this is NOT a double "newbie stick settings". Indeed, I prefer this thread, it's more "direct to the point".
As we all know, the "Formula" is a dream! I recognize. But dreams, sometimes come true. Xvid itself is a miracle of technology.It would be nice if "the formula" could work for everybody, my purpose when I wrote this topic was that we could find the most common settings, and I'm very glad people posted their valuable oppinions, I've tested some tips told here and they really improved my encoding Xvid process on low bitrates.
I think we should thank for these tips.

[snowbeach]

Encoded Taxi Driver last night. It lasts 1:53h. I am very satisfied with the rip! :D
Thumb rule (at least for me):
Movies that are not longer than 2h, you can put without problems onto 1 CD, if they are good compressable, comp test in my case sth. about 60% +! But I got really very good results, probably a good compressable movie, no noise, no filtering regarding this used. :cool:

Current XviD settings for this purpose with Koepi's build 14052003-1:
6-Ultra High
H.263
Xvid
VHQ 1 (safer for lower bitrates - below 900 kBit/s - and low quantizers - above 4)
240 (here, because of 23.976 NTSC, for more details see my guide ;) )
1
Use chroma Motion
Quarterpel
2
150
100
100
DX-50 Bvop compatible

Quantization: Leave Default!
Debug: Chroma Optimization ON
(frame drop ratio ZERO)

AND at the TwoPass tab:
Below i-frame distance...: 12 (here, because of 23.976 NTSC)
I-frame bitrate reduction: 24 (here, because of 23.976 NTSC)
bitrate payback delay: 240 (here, because of 23.976 NTSC)
payback proportionally


BTW. All these info you can find in my guide. As I said, this FORMULA already exists! :D

[bond]

Originally posted by snowbeach
VHQ 1 (safer for lower bitrates - below 900 kBit/s - and low quantizers - above 4)vhq 2-4 is already safe to use for low bitrates!!!

Quantization: Leave Default! yup you can leave it on default but you can also try 2-6-2-16 (proposed by iago)
also you can try different quantizer matrices (for example hvs-good)!

[cult]

@snowbeach
could you plz plz remove the Debug: (frame drop ratio ZERO)?
It has nothing to do with chroma optimizer.Chroma optimizer is by chance near to the option from drop ratio.It makes newbies think they can check chromaOPt and play with settings of frame drop ratio.
Plz remove it.Already have problems with some people asking me "what if I set it to.."

[dihelson]

Originally posted by snowbeach


VHQ 1 (safer for lower bitrates - below 900 kBit/s - and low quantizers - above 4)


Snowbeach, you said "VHQ 1 safer for lower bitrates", but I have seen people here advise VHQ 4 for low bitrates. Could you detail the thought line? may be today is "safer" to use VHQ 4 on low bitrates also, what do you think? Would be a bad idea?
BTW, please, I'd like to know, what soft do you use for your encodings? As a newbie I don't know "the details" :), so I use Gknot for making them. BTW, thank you for "the Guide".

[snowbeach]

@cult
you are right! i left the frame drop ration in my post, only not to confuse the people more. but i think i reached the opposite. ;)
btw. it is not included in my guide.

@dihelson
appreciate it that you like my guide. my encodings are based on my guide. the only difference is that i never encode with vhq4, because it would take me more than 24h for one encoding session on my current system. that is why i always use vhq1 and i get very nice results. i am always satisfied with the encodes.
btw. if a movie is very important to me, i go for the dvdr solution, which means that i encode it with cce (mpeg-2) and put the movie on a dvdr. best possible solution to backup a dvd. at least for me.

[EDIT]the info why it is safer to use in 1 CD rip cases vhq1, i got from koepi's build 14052003-1, when moving the mouse over the vhq option field in the global settings of the xvid settings.

[Teegedeck]

sysKin has rewritten his VHQ-code some time ago; now ALL VHQ-modes always save space AND improve quality. So you can use VHQ=4 if you want to invest the time for a better picture.

[sungey]

Originally posted by JohnMK
Before you do that, I'd suggest increasing the threshold to 100, from the default of 0. That will increase compressibility appreciably with barely any discernable quality loss, while a b-frame quantization ratio of 200 might be noticeable. In my own objective testing I found increasing the threshold to 255 barely increases compressibility over a threshold of 100, while the difference going from 0 to 100 is substantial. Giving deference to sysKin's belief that a threshold of 255 is risky, I think you can see why I'm recommending b-frame threshold 100.

In my experience, bframe threshold 255 is fine. I have seen clips where threshold 0/100/200/255 yield no difference in filesize, this probably shows that the bframe decision "knows" how to use the bframe safely :P

[vinkes]

Hi dihelson,

to improve on low bitrates, you might consider using other quantization matrices. I believe it was mentioned before in this thread, but anyway a time ago I did some testing, you can find it here:

http://forum.doom9.org/showthread.php?threadid=53594

I would definitely recommend the hvs-good matrix, as it seems to increase compressibility and not harm visual quality much.


Cheers,

[JohnMK]

This: I have seen clips where threshold 0/100/200/255 yield no difference in filesize, this probably shows that the bframe decision "knows" how to use the bframe safely :P

Doesn't logically lead to this: In my experience, bframe threshold 255 is fine.

[sungey]

what i meant was ... if bframe threshold 100 is considered safe .. and 255 is "risky" ... but in my clip both of them yield same size .. ( most likely same number of bframe in that clip, different sensitivity but same result because codec knows when not to use bframe even if the threshold is 255 ... so higher sensitivity doesnt always means more bframe ).

btw and what does "risky" means ?

[JohnMK]

sysKin feels 255 undoes a lot of other potential benefits of XviD. 0 is safe. If 100 brings all the advantages of 255, but is closer to the safe value of 0, then people should use 100.

[Acaila]

No wonder they both yield the same size, since you have max B-frames set to zero. That way there will be no B-frames in your clip anyway ;). Or was that a typo?

Besides, B-frame decision can't be smart at all with a threshold of 255, because at that setting it will force B-frames. At a setting of 100 it can be smart though.

[sungey]

@acaila
the max bframe i used was 1.

i just tested with another 30 mins movie last night, using max bframe 1 again hehehe...

threshold 0 = 298 mb
100 = 280 mb
255 = 275 mb

as expected, smaller filesize at threshold 255.

one thing that is puzzling me, we know 100 can make the encoder "smarter". But what is the advantage of this smartness ?

[JohnMK]

Well, I'd assume reduced picture quality within a fixed file size, but you should probably talk to sysKin about that. :)

[gruntle8]

I also suggest that Qpel be left off. In all my tests, I've noticed that Qpel added a large amount of noise, IMHO making the video look significantly worse than with Qpel disabled. Others have noticed this. It looks to me like with Qpel on, it's seeing a lot of moving luma noise that just doesn't exist. In some cases, it looks almost like smoke swirling over the image. Regardless, I don't see a quality gain, much less something to justify the speed and compression penalty. Perhaps in the future a fixed Qpel could offer improvements.

[kilg0r3]

Originally posted by Bilu But lesson nr.1 is: even if you kill detail below a decent level, color and movement information will allways be important players and they can stop you from fitting in 1 CD.

Hi Bilu,
does that mean that one should, in your opinion, strongly filter chroma or does it mean just the opposite?

[ominte]

I'd tend to agree with gruntle8, Qpel isn't usable at the moment because of the large amount of noise it produces. However it is undeniable that it increases the sharpeness of the video. So if you can put up with the extra noise then be sure to use it.

[bilu]

Originally posted by kilg0r3
Originally posted by Bilu

Hi Bilu,
does that mean that one should, in your opinion, strongly filter chroma or does it mean just the opposite?

I was trying to fit every kind of scene into an average bitrate of 860 Kbps (2-pass encoding) so I would be able to put 90 min on 650 MB everytime.

Try doing that with some hard to compress chapters with every kind of processing you remember and you'll see that most of the times you won't get decent results.

This was part of my learning process :)


Bilu

[thejho]

just to say that I agree with gruntle8 about qpel, I had the same problem when i tried it back in february.
Now I always turn it off.:cool:

[dihelson]

THE FORMULA UPDATED:

(Note: These are the results of compilation on the posts gathered on this Forum. Practical suggestions from people who really make Xvid, not "in thesis". )

======================================================
FOR 1 CD only:

6-Ultra High
H.263 quant
Xvid
VHQ: (Some say 1, some say 4, It's safe now to use 4, consider the amount of time versus quality).
300
1
Use chroma Motion
Quarterpel
2
150
100
100 (there's discordance on this, some recommends 100, some 255, particularly I use 255, which gave me best results.)

DX-50 Bvop compatible

Quantization: Leave Default! (2, 31, 2, 31)
Debug: Chroma Optimization ON
Frame drop ratio ZERO. (no relationship with Chroma optimization, though)

Filters ? Snowbeach and many (including me, prefers not using at all)
==============================================================


======================================================
FOR 2 or more CDs:

6-Ultra High
MPEG on 2 passes.
Xvid
VHQ: 1
300
1
Use chroma Motion
Quarterpel
2
150
100
0
DX-50 Bvop compatible

Quantization: Leave Default! (2, 31, 2, 31)
NOTE:
(I experimented by suggestions, 2,4,2,12 and the result was very nice).
Debug: Chroma Optimization ON
Frame drop ratio ZERO. (no relationship with Chroma optimization, though)

Filters ? Snowbeach and many (including me, prefers not using at all)
==============================================================

How do I do my Xvids:
Using Gknot 0.28 and VirtualDubMod basically.
I have noticed that If I use Gknot 0.28, but instead of letting it make the whole process, I use it's rip and config processes, and then SAVE the AVISynth script, and make the rest on VirtualDub for myself, the result seems better.

NOTE:
I have many times noticed that Using VirtualDubMOD cause worse Image quality. The colors seems to blend in little dots across the image,
with loss of details in dark scenes,
which don't occour using Gknot 0.27 and VirtualDub 1.4.13 instead, with the default mpeg2dec.dll (No external DLL).
The image seems to be brighter, more deep, more real, more DVD like.
It's strange this oppinion, since people often say that VirtualDubMod doesn't convert patterns of colors and could deliver a better image. That's NOT what happens here. VirtualDub 1.4.13 gives a better image instead.

* * *

[symonjfox]

Originally posted by dihelson
FOR 2 or more CDs:

6-Ultra High
MPEG on 2 passes.
Xvid
VHQ: 1
300
1
Use chroma Motion
Quarterpel
2
150
100
0
DX-50 Bvop compatible

Don't use QPel with a sharp matrix (as MPEG or Andreas78 and others). It will just create noise and artifacts.
IMHO for more than 2 CDs you can also use NO Bframes at all (depends on the source).

A general tip: since we want to create low bitrate stuff, we probably resize the source into a smaller resolution. It's important that it stays DIVIDABLE (???? :confused: ) by 32 (both vertical and horizontal). This will gain a bit the quality.

[mf]

Originally posted by symonjfox
IMHO for more than 2 CDs you can also use NO Bframes at all (depends on the source).
That is just brainless. If you have quality loss from B-frames, just do 100/100/0. That will save bitrate but keep quality.

[snowbeach]

Theory and practical testing, that is the point!

Please use B-frames and QPel and you will see that you will get beter results than not using B-frames and QPel. Believe me I tested it and I was much more satisfied with the picture I got than with the picture where B-frames and QPel were disabled. It is not true that you will get more noise using QPel, I always got reduced noise using it or simply no noise visible, or at least If you starring at the monitor and want to see some, I bet you will see some. ;) But without QPel you will see more. Trust me or try it yourself!

[EDIT] And you all know my favorite settings... :D

[symonjfox]

Originally posted by mf
That is just brainless. If you have quality loss from B-frames, just do 100/100/0. That will save bitrate but keep quality.
Thanks for the "brainless".

I heard many people that does it so. 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.

I always do 1 CD rips (with BFrames, off course), so I don't know exactly

[mf]

Originally posted by symonjfox
Thanks for the "brainless".

I heard many people that does it so. 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.
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. The only thing that might happen is that it is larger because of extra motion overhead. But considering this is a recommendation from sysKin I think that overhead will not be a factor.

[snowbeach]

You are right. Many people say not to use B-frames with high bitrates. But when do high bitrates start. My 2 CD encodes are pending between 1.200 kBit/s to 1.400 kBit/s and it is better to use B-frames, if you are dealing with bitrates in this area. Never tested it with bitrates that are higher than these, so I am not able to say, if it is better or not.

But I could imagine if you are encoding a movie to a DVD+-R/W (high :D bitrates at ~ 4.000 kBit/s) it will give you better results disabling B-frames. The other question is, if the differences are visible.

[dihelson]

Originally posted by snowbeach
You are right. Many people say not to use B-frames with high bitrates. But when do high bitrates start. My 2 CD encodes are

I agree with SnowBeach also.
IMHO, B-frames are good, I've used and like the result.
The matter of putting "0" ZERO on the threshold by making 2 CDs or more, was due to an advice by Koepi.

[mf]

Originally posted by snowbeach
But I could imagine if you are encoding a movie to a DVD+-R/W (high :D bitrates at ~ 4.000 kBit/s) it will give you better results disabling B-frames. The other question is, if the differences are visible.
If "better results" means "no undersize" -- sure. :D

[Th3-S4int]

Hi,
And how is it with trellis quant? is it good for low bitrates or not? (i ask because nobody mentioned it)