Test Results - XviD vs H.264 vs x264 vs RealVideo 10

[Seraphic-]

Well, i've spent another week testing codecs and found XviD to be the winner. I had thought H.264 or x264 would come out on top but as you can see below, that was not the case. And, as you can also see, RealVideo 10 came in last (x264 isn't much better); but I must add that RealAudio 10 has fantastic audio. I didn't test DivX as it kept giving me errors on my source video.

Anyway, I would really appreciate some suggestions on ways to improve the XviD quality. It already almost mirrors the raw avi, however, there is some color loss, blocking issues, and a few artifacts on the sharp horizontal bars. Also, could anyone suggest a MPEG-CUSTOM Matrix? I couldn't find any so I just used MPEG for the "Quantization Type" during my testing.

Comments on the results seen in the images and the settings I used are also welcome.

Thanks

RAW AVI
http://www.seraphicgate.com/junk/raw.png

------------------------------------------------------

XviD

Quantization Type: MPEG
Frame Rate: 59.94
Bitrate: 2500
Encoding Type: Twopass 2nd Pass
Everything Else: Default

http://www.seraphicgate.com/junk/xvid.png

------------------------------------------------------

QuickTime 7.1 H.264

Compression: H.264
Frame Rate: 59.94
Key Frames: Automatic
Frame Reordering: On
Data Rate: 2500
Optimized: Download
Encoding: Multi-Pass

http://www.seraphicgate.com/junk/qt7h264.png

------------------------------------------------------

X264

Compression: x264
Frame Rate: 59.94
Bitrate: 2500
Used these other settings: http://speeddemosarchive.com/kb/index.php/Megui#Setting_the_video_compression

http://www.seraphicgate.com/junk/x264.png

------------------------------------------------------

RealVideo 10

Used these below settings:
http://www.seraphicgate.com/junk/1.jpghttp://www.seraphicgate.com/junk/2.jpg

http://www.seraphicgate.com/junk/rv10.png

[iNFO-DVD]

Search for 'Sharktooth' and look at the links to his matrices. I've tried the LR and HR and they are excellent IMO. He does do some very high bitrate versions too.

[Seraphic-]

Thanks, i'll try to find those in the morning.

Here is a sample of the video in encoded with Real Video/Audio 10. As you can see, the video is extremely complex and would be hard on any codec. RealVideo 10 does an okay job with the video, but on the other hand, RealAudio 10 does a great job with the audio.

Also, isn't x264 supposed to be superior to XviD? Why did my results find it inferior?

Download Here (http://www.seraphicgate.com/junk/sample.rar) - 40MB

[foxyshadis]

Please post a clip of either of the h.264 encodes, and the frame # that screenshot was taken from. I believe you may have post-processing turned on and are seeing a bug in it (unless you're using an ffdshow-tryout from the last few weeks). Whether it is or not, you should also not blindly accept the default deblocking settings, and see if the picture improves by lowering it. Much like matrices in xvid/divx, deblocking is one of the primary determinants of picture character, and oftentimes it's a personal preference.

Oh, you're using quicktime-compatible x264 settings? Er, why? You should be making very clear that it's quicktime-compatible, ie, crippled.

Just for fairness encoding time should be listed.

Also, it really helps if you have the post ready before you post, when I first came by there was nothing here!

Off topic, from the linked site, this is kind of funny:

Number of threads - This depends on your computer. Leave it at 1 if you're cpu is single-core, and 2 if it's dual-core.
AVC profiles - Since we want to be sure that the video is Quicktime 7 compatible, you should select the main profile.

By using 2 threads you forfeit quicktime compatibility! Also this:

Number of B-Frames - I recommend 3, since it is rare that you'll see more than that in a row.

Unless quicktime's decoding has made significant progress recently.

[Seraphic-]

Hi,

The post was ready to go along with the images. However, my host (and thus the images) went down for about an hour minutes after I posted. They had to change out their core server and I didn't check the status page.

I have FFDShow [rev. 2543 & VFW interface] installed (came with the codec pack). As for post-processing, that is something i've tried to turn off many times but end up having to lower the windows "hardware acceleration" so that the videos display correctly. Maybe doing that is the reason why h.264 and x264 don't look that great.

The full video is 8 minutes and the encoding time for all of them about a hour and a half. However, quicktime 7.1 takes several hours.

Yes, I used the settings found here (http://speeddemosarchive.com/kb/index.php/Megui#Setting_the_video_compression) for x264 . Since I haven't really used x264 too much I was unsure of what settings to use. And I did not know that "quicktime-compatible" settings would cause major quality issues. Do you have a link "better" guide?

Anyway, i'll get you the other stuff in the morning! :)

[Sagittaire]

At this time your test is completely useless because:

- Source is not available for reproduce test
- Your setting are not available


From my experience for these particular source, x264 and RV10 are really better than XviD. For example for RV10 you seem not use bframe ... ???

[foxyshadis]

If you're aiming for x264 quality, well, megui does come with a number of presets. Any of the HQ profiles (decide how long you want to encode for) should be sufficient.

[Seraphic-]

I PM'ed you both a raw avi at 636x444 and 59.94 FPS.

At this time your test is completely useless because:

- Source is not available for reproduce test
- Your setting are not available


From my experience for these particular source, x264 and RV10 are really better than XviD. For example for RV10 you seem not use bframe ... ???

Don't get me wrong, I like RV10. However, as you can see in the image, xvid did a much better job. Having B-Frames unchecked means that it will be "adaptive" or so i'm told. Also, all my encode settings are listed. What settings are you talking about?

If you're aiming for x264 quality, well, megui does come with a number of presets. Any of the HQ profiles (decide how long you want to encode for) should be sufficient.

I'll try a few HQ presets, but I think I had blocking issues last time I tried. HQ-Insane has -2/-1. Frame is 161

[Manao]

Seraphic- : what are the filesizes of the different encoding ? Can I have a link to the raw too ? Care to post the settings you used ?

[Seraphic-]

I PM'ed you the link as well.

The raw file is 25.2GB and the file encoded sizes are below:
RV10 - 187MB
H.264 - 174MB
Xvid - 175MB
x264 - 178MB

What settings are you talking about?

[nurbs]

What settings are you talking about?
Ideally you should post the x264 commandline parameters you used. Otherwise screenshots like for RV10 would also do.
The guide you linked is not enough to figure out what settings you used. For instance it says use at least subme 5, but that doesn't tell us if you used 5, 6 or 7.

[Sirber]

your settings for RV10 are kinda unclear, since the GUI you used is unclear about them.

[Seraphic-]

your settings for RV10 are kinda unclear, since the GUI you used is unclear about them.

You can read about most of them here: http://forum.doom9.org/showthread.php?s=&threadid=72321

Ideally you should post the x264 commandline parameters you used. Otherwise screenshots like for RV10 would also do.
The guide you linked is not enough to figure out what settings you used. For instance it says use at least subme 5, but that doesn't tell us if you used 5, 6 or 7.

I used a base of HQ-Insane and then used the guide along with: Range of 16, Minimum Quantizer 17, Number of reference frames 5, and Number of B-Frames 3.

Would any of you like a link to the raw avi?

[Sharktooth]

25GB is too much for me.

[Seraphic-]

25GB is too much for me.

Hi,

That is the full file. I'll PM you a link to a raw avi clip that is 16 seconds. (about 394MB)

[Sirber]

You can read about most of them here: http://forum.doom9.org/showthread.php?s=&threadid=72321
I'm familiar with RV settings (I made realAnime no? ;)). All I'm saying is posting the jobfile gives the settings. Posting screenshots with 3-state checkboxs does not tell the real settings.

[Seraphic-]

I'm familiar with RV settings (I made realAnime no? ;)). All I'm saying is posting the jobfile gives the settings. Posting screenshots with 3-state checkboxs does not tell the real settings.

I was going to try your program for RV10, but I had a hard time downloading it. Anyway, I didn't know much about the settings myself. karl_lillevold, gave me some help

---------------------------------------

Here is an image from another x264 encode. I used HQ-Insane, 59.96FPS and a bitrate of 2500.

http://www.seraphicgate.com/junk/x2642.png

---------------------------------------

Also, about MeGUI. Is it true that later/newest versions seem to be missing the MP4 muxer, and the adaptive muxer does not work?

[*.mp4 guy]

Could you please compress the raw video with lagarith in yv12 mode, and send me a link to the file. I would like to see what I can do with it.

[Seraphic-]

Hi,

I PM'ed you a link to the raw avi that is already on my server. (Don't have time to upload another)

Speaking of the raw avi, has anyone else who also got a PM to the file have some test results?

Thanks

[Sharktooth]

not yet... i'll procede encoding it tomorrow.
however from your last screenshot seems the inloop deblocking settings are too high or maybe you have post-processing enabled in your decoder.

[Seraphic-]

Maybe it is the "post-processing", but I thought I turned it off (not sure if I did though). What is that even used for? Anyway, the raw avi looks fine when paused, so maybe it's off for that but not mp4 or the other formats. XviD looked okay though.

As for deblocking, HQ-Insane has -2/-1 I think. Maybe I should try 0/0.

Last, I really dislike the multiples of 16 rule. The video is 636x444 and I don't really want to resize or crop out more.

[Sharktooth]

try lower deblocking values...
like -3, -1 or -3, -2.
At that bitrate you can even try to use my EQM AVC-HR custom matrix.

[Seraphic-]

I'm only using that high of a bitrate to see some quality. I would like to use around 1750 or 2000 so that I get about 12MB per minute of video. But under 2500 and I get some really poor quality. This could be do to that damn "post-processing" though. I thought I turned it off but I end up having to lower the windows "hardware acceleration" so that the videos display correctly.

[foxyshadis]

What multiples of 16 rule? Xvid and x264 work fine at multiples of 4. No clue how strict RV10 or quicktime's requirements are though.

If all ffdshow filters are unchecked, it's not PP, it's probably just the inloop settings.

I haven't downloaded it because it's enormous, and my office's bandwidth is strained as it is (at home the dsl will simply reset every 4 minutes or so).

[DarkZell666]

Seraphic- : for your info, using lagarith or huffyuv to compress the source file will give you a smaller 3GB file instead of the uncompressed 25GB. Those codecs are lossless, so the quality will be "exactly" the same. (and you'll need 8.33x less time to upload it ;))

Thinking about the "darn post-processing" expression you used, am I wrong saying that you don't know how to use all these tools correctly yet ? No offense man, but codec comparisons don't usually take weeks or months for nothing :)

I can suggest something else which will solve the problem : use VLC to do the playback and take screenshots of VLC instead. (You're free to remove it after of course ;)) VLC doesn't have this common ffdshow PP problem.

The source is indeed difficult to encode though, so you won't find the correct settings for it in one go.

As Sharktooth said, you REALLY need to change the CQM you used with x264, high-detailed sources always look bad with the standard flat matrix. Same goes for the deblocking settings he mentionned : the default deblocking settings easily smooth out an important amount of details.

[Sirber]

I was going to try your program for RV10, but I had a hard time downloading it.My soft is ment/tweaked for 268kbps, not 2500kbps ;)

[Seraphic-]

Dark, you would be correct. I really haven't used ffdshow at all, but as you can see, PP should be off. Thus, it could be the inloop settings, but I can't seem to find it. Maybe i'll give that player a try, but it looks like it doesn't play RMVB videos. Anyway, another thing I should add is that if I do a RV10 encode at 10,000kbps it comes out looking like the raw avi. So does that mean PP and inloop settings are okay...? I don't know. I think the issue is i'm going to have to use massive bitrates to get any good quality. These videos are just too complex.

http://www.seraphicgate.com/junk/ff.png

[foxyshadis]

inloop deblocking is set on the encoder, not the decoder. It applies only to AVC video. 2500 is a very good bitrate to test this video against; perhaps 4000 would also be good.

[DarkZell666]

Hint: inloop filtering/deblocking is often referred to as "Alpha Deblocking" and "Beta Deblocking".

[killerhex]

could post those rv10 settings again the pics wont show

[DarkZell666]

+1 for the pics.

[bond]

seraphic:

just nitpicking, but x264 = H.264

[Seraphic-]

Hi Sharktooth,

Did you ever have a chance to encode a test video with x264 using that sample?

What multiples of 16 rule? Xvid and x264 work fine at multiples of 4.

When trying to use XviD or x264 MeGUI comes back with: "Clip doesn't have mod16 dimensions. This could cause problems with and will also result in loss of compressibility". Or am I able to ignore that message? (video is 636x444)

seraphic:

just nitpicking, but x264 = H.264

I listed both as I was talking about H.264 when used with QuickTime 7 and x264 when used with MeGUI.

[Sharktooth]

Not yet, sorry. I've been quite busy, more than what i expected.

[Dark Eiri]

I know little about video compression, but x264 is WAY BETTER than XviD in all means... I can encode high-res anime in 256kbps, and XviD wouldn't do that even in a dream. The same for high detailed videos.
I really don't know what is happening with your encodes.

[Seraphic-]

Not yet, sorry. I've been quite busy, more than what i expected.

Alright, no problem.

I know little about video compression, but x264 is WAY BETTER than XviD in all means... I can encode high-res anime in 256kbps, and XviD wouldn't do that even in a dream. The same for high detailed videos.
I really don't know what is happening with your encodes.

Maybe part of the issue is due to the fact that i'm having to use "ConvertToYV12()" with my test x264 encodes.

[Sharktooth]

No, it's in the nature of h.264 codecs to drop very fine "sparse" details, coz the codec acts like a denoiser.
Also some speed optimizations may come into play.

[foxyshadis]

When trying to use XviD or x264 MeGUI comes back with: "Clip doesn't have mod16 dimensions. This could cause problems with and will also result in loss of compressibility". Or am I able to ignore that message? (video is 636x444)

The message needs to be removed from MeGUI; originally it was useful when it was untested but tests have shown that it doesn't harm compressibility or quality, at least in x264. Resizing or cropping will slightly improve it just because it gets smaller, but it's not necessary anymore.

[*.mp4 guy]

Xvid Settings:
Quantization-H.263
Qpel-on
AQ-on
3 bframes - no offset, multiplier of 1.62
VHQ 4, VHQ for Bframes-on
MSP-6

X264 Settings:
subpixel refinement-6, Hex motion search, turbo enabled
All Bframes options, 3bframes, No Bprediction
High Profile, All partitions, cqm used, -2:-4 inloop, 5 mixed refs
min IDR Frame interval 48, max 240
No Fast P-Skip, and No DCT Decimation enabled

2500 Kbps:
Xvid (http://download.yousendit.com/7EB2431216C6E72C)
X264 (http://download.yousendit.com/7FB362BD10A44918)

5000 Kbps
Xvid (http://download.yousendit.com/1F2FDEE846ED7B9E)
X264 (http://download.yousendit.com/9C2F4E771ECB253E)

The files are bigger/lower quality since this sample was made from only the very dificult 999 frame sample. Xvid Files are smaller because of Xvids not-so-great ratecontrol, and my laziness in not doing multiple encodes to get it right.

[DarkZell666]

For the 2500kbps version, I can hardly distinguish which is the dirtiest, considering the blockfest that's going on :D
At least the x264 version doesn't look as washed out as Seraphic's version.
But Seraphic's encodes didn't look as blocky imho, did I not look at them properly or what ?

Edit : I can't distinguish the 5000kbps versions either .. oO

[Sharktooth]

The message needs to be removed from MeGUI; originally it was useful when it was untested but tests have shown that it doesn't harm compressibility or quality, at least in x264. Resizing or cropping will slightly improve it just because it gets smaller, but it's not necessary anymore.
No, that message should stay coz non-mod16 resolutions can cause problems with some avisynth filters.

[*.mp4 guy]

For the 2500kbps version, I can hardly distinguish which is the dirtiest, considering the blockfest that's going on :D
At least the x264 version doesn't look as washed out as Seraphic's version.
But Seraphic's encodes didn't look as blocky imho, did I not look at them properly or what ?

Edit : I can't distinguish the 5000kbps versions either .. oO

If you can't see the differences your monitor must have very low black detail, and is probably rather blurry if you can't see the difference in the blocking of the encodes.

[DarkZell666]

I must admit I'm on an 21' LCD @ work, and I'll be on a 15' laptop LCD @ home.

If anyone could give some feedback with a good ol' CRT ... ;)

[*.mp4 guy]

You may say that jokingly, but I've never seen an lcd that could match a trinitron in black depth and detail.

[DarkZell666]

Lol sorry for feeling sarcastic about the hole thing, I wasn't at all :) I know that LCD's aren't great concerning colours and lack some contrast.

My bad ^^

[bkman]

@2.5 megabits the Xvid video is blocky as hell. The x264 is of course much less blocky due to the inloop. Using nic's postprocessing filter makes the Xvid video less distracting, but the x264 vid clearly still holds more details.

[foxyshadis]

The x264 blocking most likely comes from the low inloop settings (-2,-4). I'd probably gone with -2,-1 or -2,-2 myself, to each their own.

Definitely a good video for testing against!

The only flat panels I know that can really match Trinitrons cost as much as Trinitrons did back in their day: Specialty high-res plasma monitors.

[Sharktooth]

I'd say -3, -1 or -3, -2...

[*.mp4 guy]

Just to be clear, deblocking settings are about preference, thats why most people have diffrent opinions on what looks good, even the monitor you use makes a big difference in what will look good, usually the sharper the monitor the worse high inloop settings look (but again this is still a case by case scenario).

@foxyshadis
Inloop settings are written (alpha setting) : (Beta Setting), I'm just mentioning this because you reversed the settings when you made your reply and I don't wan't things to get confusing.

@everyone who doesn't know what alpha and beta deblocking are
Alpha deblocking is the correct name for what is termed "deblocking strength" in X264, beta deblocking is the corresponding term for X264's "deblocking threshold".

[akupenguin]

I listed both as I was talking about H.264 when used with QuickTime 7 and x264 when used with MeGUI.
That's exactly what we're complaining about. What you mean is "H.264 as encoded by QuickTime and H.264 as encoded by x264". So if you want to abbreviate that, then say "h264 and h264" or "qt and x264", not "h264 and x264" nor "qt and h264".