DivX vs WMV9. Comparison of details with sample clips.

[Beastie Boy]

I set out to try and find which codec would give me the best results for high quality file archiving. Since DVD burners and discs are coming down in price, I've decided to backup my DVDs at the original resolution (no down sizing) and at a bit-rate high enough to retain as much detail as possible. This last point is important to me after seeing hi-def video and seeing what a difference resolution makes.
I carried out tests using DivX 5.1 and WMV9 VCM codecs to see which best met my needs.
In keeping with the spirit of this forum, I decided to share my results with any of you that may be interested. Take a look here. (http://homepage.ntlworld.com/paulwbeastall/details.html)
[EDIT]Beware. The page contains 2.8MB of sample images. If the buttons do not respond then the images are still downloading in the background. Please be patient.

Please remember that I carried out these tests for my own interests, please don't bombard me with a million and one different ways in which I could have done it better. However, I welcome any and all comments. Let me know your opinions.

Enjoy.

Cheers, Beastie.

PS. I don't currently use XviD or RV9 and so these were not included.

[Kalicrys]

Nice to put in a lot of different scenes for comparison.
WMV certainly does seem sharper in most of those shots.
But it also seems to exhibit a slight color shift.

[Beastie Boy]

Hi Kalicrys.
Yes, this is most noticable in the last clip on the head scarf. I'm not sure whether it is a colour shift or slight de-saturation.

[Sirber]

About the face, maybe WMV9 think it's grain.

[Soulhunter]

@Beastie Boy

Nice comparsion... THX for it !!! ;)


@Sirber
...maybe WMV9 think it's grain
LOL ! :D :D :D

[Beastie Boy]

@Sirber, I came to the same conclusion. (Sorry I didn't include RV9 :( )

@Soulhunter. No problem. I have taken plenty from these forums over the years. I thought it was time I gave something in return.

Cheers, Beastie.

[temporance]

Very interesting comparison, thanks for that :)

WMV9 must be able to save lots of bits by not encoding fine detail / noise in flat areas. It can then spend these bits sharpening up textured areas.

At least with DivX you have the choice whether to encode the fine detail. If you want it to perform more like WMV9 you just need a good adaptive noise reducing prefilter. The preprocessing built into DivX Pro is usually good for this but might be too strong for your needs even on its weakest setting.

Also, the new DivX psychovisual modes seem to enhance weaker details (e.g. textures on faces, film grain) but can causes slight softening or blocking in highly textured areas. What PV setting did you use?

When watching the videos back at normal speed, what are the most noticeable diferences?

[Beastie Boy]

Hi temporance,
PV was set to fastest, GMC and Bi-directional Encoding both used. I have updated the page to show this. Sorry, that was an omission.

When played back in real time, all the clips look very good. They have to be played back to back in order to pick out any differences.
DivX does look slightly less sharp but this is only really noticable during scenes with little movement. I'm playing the clips back at full screen with my 19" monitor set to 1280x1024. With a large hi-def display (LCD or Plasma possibly) then I think this would be more noticable, and if you use or intend to use such a display then this could be important.

The interesting thing is that WMV could easily be mistaken for the original source. The sharpness is there but the whole image is so clean that in some areas, it looks the best. The results are like a film photographers dream, sharpness without film grain. The only thing that lets it down is a slight de-saturation as mentioned earlier. But again, unless the files are played back to back you wouldn't know. The only bad clip is the one of the womans face that has been 'over-cleaned'.
During one of the high motion scenes, the original contains some slight blocking. The WMV encode also contains some blocking in the same area, but it is more even toned and therefore less noticable. I could perhaps include this clip on the web page if you like.

Cheers, Beastie.

[DigitAl56K]

Nice comparison,

I also found when testing the 5.1 beta's against WM9 that WM9 tended to lose saturation - at one point I thought I'd simply been up encoding too long and my eyes were going funny ;). For the most part its not terribly noticeable, but when explosions or fire are present in a scene it can be quite apparent.

One of my test clips at the time was the Terminator 2 title sequence. This contains a lot of rolling-fire, and when encoded with WM9 it turned from deep red to a distinctly lighter shade of orange.

I think perhaps WM9 might be saving a few bits with regards to color accuracy.

WM9 does also smooth out some light details, there is no doubt about it. Whether this is intentional or a side-effect of some algorithm that it uses I couldn't say, but simpler textures yeild lower bit-spends which in turn lower the frame quantizer and produce better quality. Getting the balance right between the quality lost by removing details and the quality gained by lowering quantizers is the hard part. Its even more complicated when you use Psycho-Vis, which attempts to control where loss or gain in detail takes place in the picture.

The PV in DivX 5.1 does indeed enhance flatter areas and mask artifacts in textured areas. The degree to which these two effects are imposed are variable by adjusting the pvLumaFlat and pvLumaTexture controls at the bottom of the Feedback window (Pro version only). Its quite interesting to pause the video during encoding, turn on the statistical overlay that monitors the PV level and play with these sliders :)

[Beastie Boy]

Yes, WMV9 certainly has some issues regarding colour accuracy. In some areas, green and blue colour channels seem over-saturated causing reds and oranges to look week.
I know I'm talking in the wrong colour space but I don't know much about YUV.

I'm hoping to upload a couple of crops from a high motion scene shortly, I,m having a little trouble with my connection at the moment.

Cheers, Beastie.

[Beastie Boy]

I have added some images showing sample crops from some fast motion scenes. (http://homepage.ntlworld.com/paulwbeastall/details.html)

Beware. The page now contains 2.8MB of sample images. If the buttons do not respond then the images are still downloading in the background. Please be patient.

Cheers, Beastie.

[temporance]

Beastie Boy,

You really should try tweaking the PV controls suggested by DigitAl56K if you're into this type of fine comparison. Maybe you'd like DivX 5.1 better with PV off. Also try setting moduation to a +ve value, and maybe try a little preprocessing. Then you'd have a codec more like WMV9.

IMHO the nice thing about DivX and xvid is that, if you want, you can configure them to behave just like WMV9.

[Beastie Boy]

IMHO the nice thing about DivX and xvid is that, if you want, you can configure them to behave just like WMV9.

Some experimentation is required I think.
I've just gotta wait for my current 22 hour encode to finish :p

Cheers, Beastie.

[IWatchHDTV]

Thanks Beastie, it's a really interesting comparison. I wonder how the results would look if you encoded a higher resolution clip. I'm surprised that Divx removed so much detail for such a small picture (688x432) at such a high bitrate. A lot of what I'm currently encoding is HDTV, and I usually resize to 1280x720, and I use a bitrate of about 2200 to make a 40 min TV show fit on 1 CD (no DVD burner :(). From the very few tests I've done, WMV retains noticeably more medium size details (like the grooves in your tree example) in still scenes, while it has less small detail (e.g., freckles) than Divx. Also, the WMV picture just looks better played in realtime)than Divx, for still scenes. However, for motion scenes, Divx has much less blocking than WMV. I guess I'm using too low a bitrate for such a high resolution. But I'm using the highest bitrate possible if I want 1 show to fit on 1 CD. Lastly, WMV 2 pass encodes about twice as fast as Divx on standard/slow 2 pass. My personal conclusion is that if you are encoding hi-res and own a DVD burner so can afford high bitrates, then go with WMV. But if you want hi-res at low bitrates (2000kbps), then Divx has less blocking.

[Lord_KiRon]

Guys what are you doing ? Even descussing M$ codecs is wrong !:devil:

Do you really want it to became standard ? With DRM and all other "stuff" it comes with ?

Even if it is MUCH better I am sure DivX and XViD will match it no time ;)

[b00zed]

Originally posted by Lord_KiRon
Guys what are you doing ? Even descussing M$ codecs is wrong !:devil:

Do you really want it to became standard ? With DRM and all other "stuff" it comes with ?

Even if it is MUCH better I am sure DivX and XViD will match it no time ;)

As much as I don't like the idea of microsoft holding standards, I was reminded of this by your post:

http://www.penny-arcade.com/view.php3?date=2002-07-22&res=l

:p

[jggimi]

Gentlemen:

Please, keep to topic, or take your discussion off line. In addition, gentle baiting can quickly degrade to flame war, and that, of course, will not be tolerated.

[SeeMoreDigital]

Originally posted by Beastie Boy
I set out to try and find which codec would give me the best results for high quality file archiving.....I've decided to backup my DVDs at the original resolution (no down sizing) and at a bit-rate high enough to retain as much detail as possible. This last point is important to me after seeing hi-def video and seeing what a difference resolution makes.
I carried out tests using DivX 5.1 and WMV9 VCM codecs to see which best met my needs. I did some 720x576 anamorphic tests at 3000kbps some time ago using DivX 5.0.5 and WMV9 VCM.

Image wise there was virtually nothing between them. However, encoding wise, WMV9 VCM took a good deal longer - at the slowest setting. But interestingly when both files were played back in the DivX player, the WMV9 encode used less processing power than the DivX encode!

When DivX 5.1 came out I began generating an encode using the 'slowest' setting (with MV) but when I worked out it was going to take over twice as long as WMV9 VCM, I abandoned the idea. No way was I going to wait 30 hours for the first pass and god knows how long for the second!

Cheers

EDIT: As for high def encoding. DivX (or XviD) is just not ready yet when encoding at frame sizes above ????x720. So you'll have to use either WMV9 (wmv), WMV9 VCM (avi) or RV9 anyway!

[DigitAl56K]

Originally posted by SeeMoreDigital
I did some 720x576 anamorphic tests at 3000kbps some time ago using DivX 5.0.5 and WMV9 VCM.

Image wise there was virtually nothing between them. However, encoding wise, WMV9 VCM took a good deal longer - at the slowest setting. But interestingly when both files were played back in the DivX player, the WMV9 encode used less processing power than the DivX encode!


DivX 5.1.1 Beta 1 is now available and is 50% faster than 5.1. The GUI is now "locked down" to some extent to prevent encoding boo-boo's.

The DivX 5.1.1 Beta 1 decoder is also at least 30% faster than the DivX 5.05 decoder and features real-time adaptive post-processing, so that by default the maximum post-processing supported by your CPU will be used while taking care not to drop frames by allowing the CPU to become overloaded.


When DivX 5.1 came out I began generating an encode using the 'slowest' setting (with MV) but when I worked out it was going to take over twice as long as WMV9 VCM, I abandoned the idea. No way was I going to wait 30 hours for the first pass and god knows how long for the second!


Try 5.1.1 Beta 1. Do your first pass in Standard mode, then one nth pass in either Slow or Slowest mode. Don't use the MV file.


EDIT: As for high def encoding. DivX (or XviD) is just not ready yet when encoding at frame sizes above ????x720. So you'll have to use either WMV9 (wmv), WMV9 VCM (avi) or RV9 anyway!

Its been ready for high-def since 5.0, with high-def profile support since 5.03 (though I would recommend a minium of 5.05).

[Beastie Boy]

To be honest, encoding time is less important to me than the quality of the output.
I am currently encoding a 1.85 ratio anamorphic movie with no cropping, 720x576 res using WMV9 VCM, Complex setting, 1 notch down from slowest. I estimate that this will take 30 hours.
I find this is marginally faster than DivX 3 pass, 2@standard speed, final @ slowest speed. I did the comparison purely to see which gave the best looking output, regardless of encode time.

Cheers, Beastie.

[SeeMoreDigital]

Originally posted by Beastie Boy
To be honest, encoding time is less important to me than the quality of the output.
I am currently encoding a 1.85 ratio anamorphic movie with no cropping, 720x576 res using WMV9 VCM, Complex setting, 1 notch down from slowest. I estimate that this will take 30 hours.
I find this is marginally faster than DivX 3 pass, 2@standard speed, final @ slowest speed. I did the comparison purely to see which gave the best looking output, regardless of encode time.

Cheers, Beastie. Personally, I found that when encoding using WMV9 VCM at full bar (slowest), what goes on in the background of the overall image, become much more stable than when using WMV9 VCM at 1 notch down from the slowest.

Admittedly my WMV9 VCM 'video only' tests were done using a bitrate of just 617kbps (StarWars 2 Chapter 41 720x576 pixels 'anamorphic'), which is probably a lot lower than you will be using. My DivX 5.0.5 'video only' tests were done at 627kbps.

I'm generating some new tests today using DivX 5.1.1 beta1, which will be at 627kbps too. At least this way I can compare the new beta tests against old DivX 5.0.5 tests!

Cheers