DivX5's quality/filesize claims

[JimmyBarnes]

Call me perverse but I greatly prefer to do single CD rips.

DivX claims to produce comparable quality to DivX4 but in a 40 % smaller filesize. Quite a claim, and if true should make quality rips possible of even 3 hr movies on a single CD.

Has anyone done significant numbers of DivX4 and DivX5 rips to say whether this claim appears reasonable or not?

JB

[N_F]

While I have not tested exactly what you describe, I can say that it is not a reasonable claim. Perhaps under "perfect" circumstances where DivX5 would produce great results and DivX4 bad results it would be a reasonable claim, but AFAIK there isn't such a circumstance.

The way I see it DivX5 is definately better, but not nearly as good as in 40% smaller filesize without much loss in quality. Perhaps 5% if you really want to put a number to it?

[JimmyBarnes]

Originally posted by N_F
The way I see it DivX5 is definately better, but not nearly as good as in 40% smaller filesize without much loss in quality. Perhaps 5% if you really want to put a number to it?

Pity if it really is that small. Thanks for your reply but I guess I'll have to wait for other responses as well.

I've heard some saying they use a 1st Pass % figure of as low as 50 % when using DivX5. I'll be glad when Gordian Knot catches up with DivX5, especially with respect to filesize prediction for a suitable resolution/quality.

JB

[DJ Bobo]

@ N_F
You're kidding! there is 20 to 30% better quality using the PRO Version!

It's simple to test that:
Encode the same sequence in quality based 100%:
with DivX 4.11
with DivX 5.01 PRO without MPEG4 tools
with DivX 5.01 PRO with GMC & Bidir. Encoding

And you'll see that the last version will produce a much smaller file!

[N_F]

Originally posted by bobotns
@ N_F
You're kidding! there is 20 to 30% better quality using the PRO Version!

It's simple to test that:
Encode the same sequence in quality based 100%:
with DivX 4.11
with DivX 5.01 PRO without MPEG4 tools
with DivX 5.01 PRO with GMC & Bidir. Encoding

And you'll see that the last version will produce a much smaller file!


Well that wasn't really what I meaned. The way I see it, to compare two codecs you should use realistic settings. Quality based 100% is not realistic. You may be right in what you're saying, I wouldn't know about that, but again, it is not realistic settings and wouldn't make a difference to the "normal" users.

If you do a 2-pass encoding at 800 kbit/s (quite realistic) with DivX 4.12 and DivX 5.0 (haven't tried 5.01 yet, but it shouldn't be that big a difference) there will be a difference, but no way like 20% to 30%.


At least this is my experience with the two codecs... Would be interesting to see some more input.

[Sky]

With 100% quality based, most of the movies take place on 2 CDs incl.
AC3 if they have a 1:2.35 ascept ratio and at 1:1.85 its no problem
with MP3 Audio.
I prefer 2 CDs with really good picture qualitity at 480x384
and not as many people here a higher resolution on one Disc.
Because I use TV-Out a resolution higher than 480x384 is really
recognizable.

[Acaila]

If you do a 2-pass encoding at 800 kbit/s (quite realistic) with DivX 4.12 and DivX 5.0 (haven't tried 5.01 yet, but it shouldn't be that big a difference) there will be a difference, but no way like 20% to 30%.

When you use 2-pass mode the codec will use whatever quants neccessary to achieve the bitrate you set. This means you will see NO difference in output size, because 800kbit/s remains 800kbit/s no matter what codec you use. The 800 kbit/s determines your output size, and output size is thus fixed.
The only thing you can check with your method is which codec achieves higher quality at equal filesize, and such a test is highly subjective.

Using bobotns' method you will see how much size the codecs need to achieve the SAME quality. Thus lower filesize is better. This is a lot less subjective because you don't have to judge for yourself.


I prefer 2 CDs with really good picture qualitity at 480x384
and not as many people here a higher resolution on one Disc.
Because I use TV-Out a resolution higher than 480x384 is really
recognizable.
I never watch my movies on TV, but do you really need that low a res and for 2 CD's even to get an excellent picture when watched on TV?
For watching on a PC monitor even 640 on 1 CD gives a perfect movie. Or 704 for 2 CD's.

[jonny]

Make a compressibility test to see how the new options increase compressibility... b-frames really make the difference

[ulfschack]

I'm with Jonny on this one.

Use his tool to get it working in Gknot. Word is that the Codec saturates on 80% it seems, so using this as threshold you could calculate the difference for the corresponding bitrate in DivX4. That way we can get pretty reliable "numbers on quality" and dont have to rely on peoples different views on quality.

This of course assumes that Dix4 and 5 regards "quality_sat" as being the same in both instances. :o

[Sestrata]

Jimmy, I have done 18 movies with 4.12 and 8 with 5.0. All 640x532(272). Believe me, 5.0 is far ahead, not only at size. It was visible, that 4.12 makes small objects in background to move and lose contrast. 5.0 is many steps better! 3 hors movies like "Mists of Avalon" and "Pearl HArbor" are perfect now. I use all three extras + force max quant 2. With DivX 5.0-01, 650 kb/s for video is still acceptable.
By the way... why everybody burned at 650-700 mb?! For example: eximpo-crd.cz sells 99 min blanks with 870-900 mb for only 0.71 DM/pcs. Only old dvd-roms don't read them. So why to encode at 2 CD???

Sestrata

[DJ Bobo]

@ N_F
I think Acaila gave a good answer.
If you allow, I'll add another thing: we keep the example of 800kbps 2-pass encoding.
If we have for example 25% less size with quality based 100%, that means, when we do 2-pass encoding @ 800kbps with GMC & Bidir. Encoding, we get the same quality as 2-pass encoding @ 1060kbps without GMC & B-E ;)
And trust me, quality based 100% is also realistic, I got some movies that were very good compressible and got them with quality based 100% and 2 audio streams on 2 CDs my friend ;)
For example: Eraser (quality based 100% with GMC & B-E)704x304 25fps with AC3 @384kbps + MP3 @160kbps on 1,3GB (2 74-min CDs)
Or Coyote Ugly (quality based 100% without mpeg4 tools)704x304 25fps with AC3 @384kbps + MP3 @144kbps on 1,5GB (2 90-min CDs)

@ Sestrata
That if the burner is compatible with them.
99-min CD-R are not widely availabe because their compatibility is very bad.
90-min CD-R are quite available, because their compatibility is a lot better. But many drives can't read them, I have a lot of friends that have drives that can't read them.
Well in my case, I can read them and burn till 89:30, but not more.

[N_F]

Originally posted by Acaila


When you use 2-pass mode the codec will use whatever quants neccessary to achieve the bitrate you set. This means you will see NO difference in output size, because 800kbit/s remains 800kbit/s no matter what codec you use. The 800 kbit/s determines your output size, and output size is thus fixed.
The only thing you can check with your method is which codec achieves higher quality at equal filesize, and such a test is highly subjective.

Using bobotns' method you will see how much size the codecs need to achieve the SAME quality. Thus lower filesize is better. This is a lot less subjective because you don't have to judge for yourself.

Ok, I now understand why doing a comparesion (sp?) at CQ 100% isn't a bad idea. But I still think that while such test as I mentioned would be subjective, if DivX5 is so much better as some of you seem to think, the difference in quality would be so big that it doesn't matter that is is a subjective opinion. Of course it's very hard to put a percentance on how much better DivX5 is in a test like this.


Originally posted by bobotns


If we have for example 25% less size with quality based 100%, that means, when we do 2-pass encoding @ 800kbps with GMC & Bidir. Encoding, we get the same quality as 2-pass encoding @ 1060kbps without GMC & B-E


Ok, so what you're saying is that DivX5 800 kbps GMC & B-frames will produce similar results (in quality, not in size) as DivX4 ~1060 kbps?

I'm sorry but I find that very unlikely. I don't mean to be rude or doubt you, I just haven't seen what you're saying. I will test these specific settings during the weekend. Maybe you're right. :)

[theReal]

I'm with Jonny on this one: make a compress. test in Gordian Knot. You will see that if Divx5 without any of the pro settings is close to 50%, you will come close to 60% when you enable b-frames (and the result is of the same quality, in my opinion).

In addition to that, I also think that Divx5 has a better visual quality than Divx4, it's a little more detailed.

So, Divx5 doesn't produce a much superior quality than Divx4, but it's a little better. At the same time you can make files 20-30% smaller when using b-frames (if it still has the same quality then is debatable - I think so!)

[Captain_Nipple]

Hey,
I have found that if i use GKnot and get a DivX4 bitrate(1kBit=1000Bit) and then use that result in DivX5.x, it would stick to the final filesize that GKnot found regardless of the extra features in DivX5.x (e.g if they were enabled or not)
I have done many long dvd this way.
Godfather trilogy, Heat, Goodfellas to name a few. All quite nice one cd rips.
The filesize would be sometimes -+1mb, which I think is quite good.
Just my experiences.

Captain

[theReal]

@Captain_Nipple,
this is my experience as well. Divx5 sticks to the bitrate very closely (better than Divx4). The difference is, with b-frames you can use a smaller bitrate and achieve the same quality as with a higher bitrate without b-frames.

[TheXung]

if DivX5 is so much better as some of you seem to think, the difference in quality would be so big that it doesn't matter that is is a subjective opinion. Of course it's very hard to put a percentance on how much better DivX5 is in a test like this.

If you make a 2 CD rip and a 1CD rip, is the 2 CD rip double the quality? No. You're assuming there is a linear relationship between quality and filesize when it fact there isn't. The claim was on filesize, not quality. I don't think you have enough experience doing 2 CD rips and seeing the quality you get from adding bitrate.

The other misconception I see is that DivX5 and DivX5 with bframes are not equal in quality. It is almost equal but not quite. It's filesize claim is made assuming that turning on bframes leaves the same quality. All in all, it's not a 40% improvement, maybe 20-30%.

[N_F]

Originally posted by TheXung


If you make a 2 CD rip and a 1CD rip, is the 2 CD rip double the quality? No. You're assuming there is a linear relationship between quality and filesize when it fact there isn't. The claim was on filesize, not quality. I don't think you have enough experience doing 2 CD rips and seeing the quality you get from adding bitrate.

The other misconception I see is that DivX5 and DivX5 with bframes are not equal in quality. It is almost equal but not quite. It's filesize claim is made assuming that turning on bframes leaves the same quality. All in all, it's not a 40% improvement, maybe 20-30%.


Does filesize = quality ? When the encoding is done right and at reasonable bitrates (for 1CD!), YES. The general answer would be no since after ~800 (may be 700, may be 900, I don't know that specific) kbps there isn't a linear relationship between quality and filesize as you have already mentioned.

Did I say that DivX5 and DivX5 with B-frames produce equal quality? I can't remember that, but if I did, that was a misconception as you mention.

I've been running tests on this during the weekend and still are. Should be done tonight.

A quick thought: 100% CQ with B frames gives a filesize about 25% lesser than the same settings without B frames. But are these two clips equaly good in quality? If they are, then B frames should make a very big difference. The thing is, I don't really think the clips are similar in quality. My eye can't see any difference, but that's because the difference probably is very small (again, the relationship between filesize and quality comes into question).

I'll report my results more specific later.

[theReal]

My eye can't see any difference, but that's because the difference probably is very small

That's exactly what I'm thinking: the quality is definitely worse with b-frames, because b-frames use a quant multiplier of 2.0000 (you can see this in the log files). Therefore, a quant 2 encoding (=100% quality) uses quant 4 for every b-frame. That is technically, theoretically and mathematically worse than using quant 2 for every frame. But do I see it? I don't, so I don't care and I'm happy about the smaller filesize :)

[oddball]

I don't understand. Smaller filesize? If you encode at 800kbit with or without B frames it should still end up the same filesize no?

[oddball]

Or rather it should be the same filesize but with B frames enabled should be of better quality because it handles the bits better.

[theReal]

I don't understand. Smaller filesize? If you encode at 800kbit with or without B frames it should still end up the same filesize no?Yes, but I was talking of quality-based (=fixed quantizer) encoding.

And yes, because it handles the information better, b-frames can be smaller at the same quality - there's no option to make better quality videos with full-sized b-frames.

[dragoman]

Hi,

I think you are missing the point of b-frames:

there's no option to make better quality videos with full-sized b-frames.

What I think you are saying here is that you wish to be able to set the quantizer of the b-frame to the lowest possible if you wish.

There is no need for this option because if you encode at 100% quality based and turn off b-frames it will accomplish the same thing.

B-frames are not designed for this: they are designed to make video the same quality at a smaller filesize. If you don't care about filesize, only quality, then don't use them. For you, there would be no point.

dragoman

[oddball]

So if you want to create a 700MB file with the best quality what would be the correct settings? Doom9 says to put B frames on in his guide no?

[N_F]

oddball: Pretty much everyone seems to agree B-frames should be turned on for the best quality.

Something I don't quite understand is the following: When I encode the same clip with the same bitrate both with and without B-frames, the size of the file is pretty much the same. When I activate B-frames, do DivX automatically add more bits to the picture to compensate for the lower filesize B-frames should generate?

[ulfschack]

@N_F: This has been said before. Given that the complexity of the video is so that the bitrate doesn't saturate that quality it's only logical that they end up the same size. This is what (average) bitrate is good for.

However, the b-version should (and will) hold better (visual) quality.

cheers