Here we go ! First real test is starting

[Actron]

Ok guys....I dont have anything to do, so I will start now the first quality test of divx 5 pro and xvid 2002.03.02 !

the film im going to encode is "A Knights Tale" (german dvd version, about 2h long)

there will be :
-1cd xvid version
-2cd xvid version

-1cd divx5 pro version
-2cd divx5 pro version

probably more to come but thats it so far.

I will try to make the best settings for each codec (no prob with xvid, im pretty experienced with that, but divx5 is new....)

screenshots and personal opinion will be here in about 12-16 hours :)


greetings

actron

[Actron]

1 cd xvid and 1 cd divx 5 versions are now finished, but i dont have the time right now, to compare them with each other...damn...second pass using this dv file is really fast with divx5 :)

[Actron]

Hey guys, the comparsion is now finished !!!

please visit it, dont complain, be friedly and just look at the pictures, although they are very big ! :)

actron

http://www.netcologne.de/~nc-varnhaga2/comparsion/comparsion.html

[Actron]

hmmm....no comments ? :(

[Zarxrax]

Very interesting. At first I was really excited about divx5, but after this and my own tests (see threads http://forum.doom9.org/showthread.php?s=&threadid=19313 & http://forum.doom9.org/showthread.php?s=&threadid=19564)
I am now just about convinced that Divx5 is pure and utter crap.

[Actron]

ahhh, a comment :)

yes, thats the exact same thing, I found out myself....But how can it be, that DivX is so bad ??? anything must be wrong....

[dado023]

judging from the pictures divx5 is worse than divx4
are you sure you set up everything correct in divx5 settings
a have done a few movies at low bitrates like this in DIVX4 and i didnt ever get those blocky results.....

btw:in wich resolution did you encode

[rui]

Well, I just finished my The Replacements 1 cd rip, using both Divx5 Pro and Xvid, latest Koepi’s build.
For divx5 I enabled all the 3 MPEG4 options (quarter pixel, GMC, and B-frames), no pre processing options or psychovisual enhancements.
For xvid, I used 40% key frame boost, 25%/15% high and low curve compression and payback delay 240. Both intra and inter frames were at their defaults. No lumi masking and the quantization was H.263 for first pass and modulated for second pass.
I used in xvid the credits option, using a 31 quantizer for both inter and intra frames.
The final file size with audio was 700 MB for xvid, and 703 MB for divx5 (I was very surprised to see that divx5 almost it the desired file size). I used Gnot to obtain the bitrate that I placed in the divx5 codec bitrate tab.
To see both avi’s, I used for xvid the latest Nic’s post processing filter with all the options enabled. For divx5 I used a post processing value of 2.

And now for the results: none of the two codecs could give me a good 1 cd rip. This movie still has to be beaten. (In 1 cd rip, I mean).

Both avi’s were very bad in the initial underwater scenes.
Both the avi’s had a kind of glow along the edges. This was particular noticeable when the scene was a close-up of a person. But xvid was better here. Divx5 still has this problem, which is a heritage from divx4. Xvid colors seem a little more “real” than divx5. Also, I think xvid can handle fast scenes better than divx5. When the players moved around the field, in the divx5 rip one can notice a little artefacts around the person.

But, now the great surprise, at least to me: divx5 is, as of today, sharper than xvid. I must be honest here and say it. The divx5 rip retained more details (I already read some posts saying this, and I must agree with those people) from the DVD than xvid.
Maybe this was caused because, even by choosing the modulated option, I could see in debug view that most of the time xvid was using H.263, which is know to bee more blurrier than MPEG.

But, we must remember that divx5 already has B-frames, and xvid not. So, when xvid gets b-frames, I am certain the quality will improve, and then my conclusions could be very different.

I decided to post this both in divx5 forum and xvid forum, in spite of the warning made by doom9 about comparisons of codecs should only be posted in the divx5 forum. This respects both xvid and dvix5, so I think is justifiable to post on both forums.

And finally, in spite of the fact that doom9 advised that codec comparisons should be made using quality passes for both, I don’t agree. I think that what is most important for the user is the quality each codec can attain by respecting a certain file size. So I decided to do this test using 2 passes to reach a certain file size.

Actrom, hope you don’t mind to use your thread to post this. I saw that you came to somewhat different conclusions, but that only proves that each one should do his tests and then judge for himself.


P.S. Sorry, i don't have any pictures to show you guys

[canadian_fbi]

Originally posted by rui
And finally, in spite of the fact that doom9 advised that codec comparisons should be made using quality passes for both, I don’t agree. I think that what is most important for the user is the quality each codec can attain by respecting a certain file size. So I decided to do this test using 2 passes to reach a certain file size.
i have to say i agree 100% with this point. for tests of the codec's ability to reduce filesize, or perhaps testing the effects of the various options of divx 5, constant quality is the way to go. however in a comprehensive "real-world" test like seeing how a whole movie will encode, how many people are going to do this using a constant quality or quantizer setting?

unless i'm mistaken, 2-pass is the preferred method for sticking to a certain filesize, so any problems or shortfalls the divx 5 codec has with 2-pass encoding are extremely relevant. (and i've seen some posts that imply that advantages that divx 5 has in 1-pass quality mode get wiped out in 2-pass mode.) even if divx 5 beats xvid in a constant quality test, if xvid wins in 2-pass mode, i'm using xvid to encode movies with 2-pass.

so don't get me wrong, constant quality tests have their place (like to determine that quarter pel doesn't do much), but tests that carry the most weight with me are 2-pass tests of full movies of the sort actron and rui have just done. and i must say that their results are quite surprising and also very much appreciated :)

[Actron]

Actron, hope you don’t mind to use your thread to post this. I saw that you came to somewhat different conclusions, but that only proves that each one should do his tests and then judge for himself.

No, prob, but we have more or less the same conclusion I think ! I also say, that DivX 5 has more details, than Xvid (also have a sample pic for that.).

Hmm...After all, I think DivX 5 is not that good, it tries to be..not more not less....

[AM Tiainen]

Use Quarter Pixel. I know that doom9 say this:
Use Quarter Pixel: Using this setting will result in a resolution of a quarter pixel for the motion compensation. This allows a better resolution for the motion compensation as a block may sometimes move less than a pixel. I've been told by the developers that this feature only seems to work well when you quarter the size of the original video. I've tested it at the standard resolution of 640x and found that it creates weird effects so you should disable it.

But that is not true (may that can cause that problem in few machine). Try youself. Doom9 don't recommend that but i don't see any reason why not try that options.

Also "rui" say that quality was very same in both codecs. In your test DivX 5 was very poor quality, and that cause because you don't use Quarter Pixel (or so i believe).

[3D]

I have been running some tests as well. With the source that I have used, The Score, I received a 100mb size savings from 4.12 at the same bitrate. Pics here (http://www.dandirksen.com/divx5.asp). I used only the b-frames option ans the same avs for both encodes. Last night I attempted to "use" the saved space by encoding at 1400 kb/s but I selected the wrong avs (it has the credits and a larger resolution). The higher bitrate did not solve macro blocks in one scene and the file was oversized (probably because the credits took a lot of bits). Will run the higher bitrate on the same avs as the previoud attempts to see if that helps, but I think that both 4.12 and 5.0 have the same "block" problem with shadows and solid colors as we all have seen. But at the same time one cannot dismiss the file size reduction that I have seen, as at the same bitrate the quality was the exact same, the macroblocks are even almost identical...

[Actron]

In your test DivX 5 was very poor quality, and that cause because you don't use Quarter Pixel (or so i believe).

I didnt use it because it read in the forum that it can blow up your framesize and doesnt give much improvement...(by the way, it is said to dramaticle decrease your encoding speed, and since i had to do 4 full dvds encodings in one night for this test :) )

BUT :

Can anyone confirme, that Quarter Pixel gives much better quality ?

actron

[rui]

Originally posted by Actron


I didnt use it because it read in the forum that it can blow up your framesize and doesnt give much improvement...(by the way, it is said to dramaticle decrease your encoding speed, and since i had to do 4 full dvds encodings in one night for this test :) )

BUT :

Can anyone confirme, that Quarter Pixel gives much better quality ?

actron

Actually...

In my first test I used the quarter pixel, like I said. I used a resolution of 576x320.

And I did experience some artefacts. I didn’t mention it in my post because I already had read the other posts stating that using quarter pixel could cause problems. So, I figured out that I shouldn’t mention it because it was already recognized that that feature could cause problems.

I made two quick tests using The Replacements trailer, using the same resolution indicated above, one with quarter pixel and the other without it, and the second one has no artefacts. The one with quarter pixel has a little artefacts, just like my full movie test.

So my advice is NOT to use quarter pixel, at least for now. Or if you want it, I suggest making a quick test using the same resolution that you plan to encode your full movie, and see if there are any artifacts.

For the fact of not having mentioned the artifacts in the first place, i apologize to everybody :(

[Actron]

im very sorry, but my comparsion if offline, because my transfer limit is heavily exceeded....oh man, 16 thousend hits in a few days :)

# Hits: 15874
# Bytes: 2637417253
# Megabytes: 2464.70302009583
# (1 MB = 1024 kB, 1 kB = 1024 bytes)


pretty cool ! :)

greetings

actron

[AM Tiainen]

Originally posted by Actron

Can anyone confirme, that Quarter Pixel gives much better quality ?

actron

Yes.
Test to see DivX 5 options and difference is http://mukoli.free.fr/video/ and you can see clearly that Quarte Pixel options make better quality.

You can view MPEG-4 codecs comparisor from http://www.nixed-hq.com and press "Codec Comparisor" link
or directly http://koti.mbnet.fi/website/mpeg4test/ . In there test DivX 5 is make with all MPEG4 options and you can notice how good quality it make when use all DivX 5 pro options (Quarter Pixel + GMC + Bidirectional Frame).