Anyone used WM9 encoder for any encoding lately?

[31 Flavas]

Originally posted by Neo Neko

Or Sega making a Nintendo game.Hehe, except, ... Sega DOES make Nintendo games!

Super Monkey Ball 1 and 2 and NBA 2k3 and the Sonic Mega Collection game (if you need some sonic goodness) to name a few GameCube titles... :)

Come on Microsoft, you can do it!

[RadicalEd]

Neo knows that, it was to put irony in what amir said about sony and nintendo ^_^;

[31 Flavas]

Originally posted by RadicalEd

Neo knows that, it was to put irony in what amir said about sony and nintendo ^_^; Ahh, I should have known! :) <smacks forhead>

[amirm]

Originally posted by Neo Neko
Here is a question Amir might be able to answer though. I have heard from people I trust that at least on internal code there are switches that can be used that cause WMV9 to output ISO-MPEG4 streams. Is this true. If so is it something we might see though possibly not publicised on the final release version?

There is no way to get the WMV 9 codec (or 8) to produce MPEG-4 compliant bitstreams. The two technologies are so different that you basically need the code for each to produce such a result. This may have come from our prior variations of MPEG-4 where we could indeed, generate MPEG-4 compliant streams. But this is no longer true.

While I am typing this, here are some remarks on your other comments:

1. My feelings are not hurt at all and I don't take any comments personally in this forum. And I very much appreciate the welcome.

2. We did not "remove" the filter properties from Media Player 9. The 7.X through 9.0 media players do not use Dshow for their core as the 6.4 player did. The 6.4 player in many respects was just a simple UI on top of the filters so it exposed a lot more of that functionality. People miss the 6.4 player sometimes but I must say, this is the first time I have anyone miss the filter properties :).

3. You say that On2 tried to keep up with WMV 8. WMV 9 outperforms V8 from 15 to 50%, depending on content. Having gone through the exercise of making WMV 9 better than V8, I have to say that it was a much harder job than outperforming less capable codecs such as MPEG-2/MPEG-4. So anyone who thinks they can make On2 better, has their work cut out for them.

4. I see a lot of quality claims wrt to WMA and WMV but it is very hard for me to address them without seeing the clips, encoding parameters, tools used, etc., etc. One wrong parameter and you have apples and oranges comparison. The devil is always in the details.

I implore everyone to do their own testing and not draw quick conclusions from casual tests. WMV has been compared and tested by experts alike and it always, and I mean always outperforms MPEG-4 derived codecs, not matter what their origin is (Divx,, etc.).

5. I keep seeing the comment that WMV is a "low-end" codec because we do well at streaming rates. But this is an improper generalization. WMV is scales very nicely with data rate and resolution because it is optimized differently for various "profiles" (another reason to not generalize its quality from encoding at one or two data rates). As a matter of fact, most of the kudos we receive are at the higher end of the spectrum. Here are some examples:

We have people projecting HD WMV on 30+ foot screens at data rates that are the same as standard resolution MPEG-2 on DVDs. See BMW Films Release (http://www.microsoft.com/presspass/press/2002/Oct02/10-30MotownPR.asp). And people can't tell the difference from film in surveys done after the viewing. There is tremendous amount of film grain in the "Shadows" movie and all of it is preserved like no other codec can. Sundance film festival just added WMV as one of its approved submission formats for independent films. No other compressed format is allowed (all the other formats are broadcast quality tape and 16/32-mm film).

Here is a super long thread on avsforum on people encoding with beta version of WMV 9 their 1080p content. http://www.avsforum.com/avs-vb/showthread.php?threadid=171590&perpage=20&pagenumber=1

Here are some comments on the quality:

"Amazing. On my 2.8Ghz 1Gb machine it plays well Simply amazing quality"

"All I have to say is "Wow!" .. I'm looking very enthusiastically towards HDTV"

"<In the voice of Janice from Friends> Oh, my god! </In the voice of Janice from Friends>

That rocks! Now there's no turning back, either I have to emmigrate to the states or get me a HD pimp."

You get the picture I think :). BTW, this is where I ran into Nic.

Finally, note that we have made some incremental quality improvements from Beta to RC so the results should be even better (and especially so if the slider on the Encoder is pushed all the way to the right).

Amir
Microsoft

[TheXung]

From reading Amirm's posts, I get the impression that he sees the alternative codecs (real, on2, wmv8) as above the mpeg4 ones whereas around here we all have concluded that the alternative codecs do not hold their own against a well tuned mpeg4 codec like xvid or divx. I wonder where this inconsistency comes from. We have either become too "use" to the look and artifacts of divx and xvid too make an objective conclusion or companies are refusing to treat these "outlawed" codecs with a fair shake, turning a blind eye to the development of these outlawed codecs. (I am awared that Divx today is not illegal, just that many businesses still associate the DivX name with an illegal codec).


@Amirm
What sort of clips are you refering to when you make quality statements about WM9? It is not like we can just download the WM9 demoes, reencode them using divx and then make a fair comparison; we would need the originals that they were encoded from. I have done my own testing and the quality claim doesn't hold up. The only thing I can think of that can make wm9 better is that in a variable quality encode, the higher quant scenes don't need as many bits to match divx and so can devote more bits to the low quant scenes. But when I try out a codec, I first look at how it performs at a given quality and not at it's bitrate controling or multi-pass compression. Those are only further investigated if I am satisfied with the strength of the encoder.

Perhaps the presence of any macroblocks really disturbs people and we have just gotten use to them for fast action scenes. Maybe that is why all of the alternate codecs choose to produce a softer image.

[trbarry]

I would love to see a high bit rate codec shootout. ;)

- Tom

[Sgt_Strider]

Originally posted by trbarry
I would love to see a high bit rate codec shootout. ;)

- Tom

That would be nice :).

[temporance]

Originally posted by amirm
As a matter of fact, most of the kudos we receive are at the higher end of the spectrum. Here are some examples:

We have people projecting HD WMV on 30+ foot screens at data rates that are the same as standard resolution MPEG-2 on DVDs. See BMW Films Release (http://www.microsoft.com/presspass/press/2002/Oct02/10-30MotownPR.asp). And people can't tell the difference from film in surveys done after the viewing. There is tremendous amount of film grain in the "Shadows" movie and all of it is preserved like no other codec can.
Amir, what was the actual bitrate? AFAIK, DVD bitrate can be as high as 15Mbit/s. Film grain (additive noise) cannot be compressed at all at lower resolutions (entropy - it looks like white noise), but a HDTV scan of small format fast celluloid (e.g. 16mm) would produce coarser grain and hence more compressible material.

Sundance film festival just added WMV as one of its approved submission formats for independent films. No other compressed format is allowed (all the other formats are broadcast quality tape and 16/32-mm film).
Apologies for appearing cynical, but would that be the Sundance film festival that is sponsored by Microsoft?

Here is a super long thread on avsforum on people encoding with beta version of WMV 9 their 1080p content. http://www.avsforum.com/avs-vb/showthread.php?threadid=171590&perpage=20&pagenumber=1

Here are some comments on the quality:
...

These people are enthusing because they have never seen HDTV before, not because it's WMV9. What is being done here is reencoding from MPEG-2 using a more modern codec. (Sound familiar, doom9 readers?) And the new bitrate is approximately half the ATSC MPEG-2 bitrate (max 19Mbit/s, typically 10-12Mbit/s for 720p).

Finally, note that we have made some incremental quality improvements from Beta to RC so the results should be even better ...
Look forward to seeing the RC. And would be even more interested in two things:
A technical explanation of why WMV9 should be better than, say, MPEG-4 - what are the extra tools that improve the quality (we know it has qpel and bframes).
An independent shootout done by someone from the doom9 pages (doom9 himself?)

Thanks Amir for your contributions here - I'm sure you must be busy with your upcoming release, hope you stay around.

[Nic]

Ive updated the WM9Encoder on my site with the latest (i think) codec ( http://nic.dnsalias.com )

Re-download the codec as well, not just the installer.

Cheers,
-Nic

[Rash]

Originally posted by amirm
While you and I might like the Dshow filter properties feature of 6.4 player, 99.999% of our customers wouldn’t know what a Dshow filter is let alone want to set its properties :). So for good or bad, the masses do not have the same requirements or preferences that you state.

Amir, just a doubt. Is it possible to make a basic and advanced codec at the same time. I mean, something that can be useful for the masses that don't know DShow and to people that know what it is and like playing with it? :)

[Neo Neko]

Originally posted by amirm
There is no way to get the WMV 9 codec (or 8) to produce MPEG-4 compliant bitstreams. The two technologies are so different that you basically need the code for each to produce such a result. This may have come from our prior variations of MPEG-4 where we could indeed, generate MPEG-4 compliant streams. But this is no longer true.

Too bad. Now that you mention Nic I think he was the one I had heard that from. ;)

Originally posted by amirm
2. We did not "remove" the filter properties from Media Player 9. The 7.X through 9.0 media players do not use Dshow for their core as the 6.4 player did. The 6.4 player in many respects was just a simple UI on top of the filters so it exposed a lot more of that functionality. People miss the 6.4 player sometimes but I must say, this is the first time I have anyone miss the filter properties :).

This is the first time you have ever heard of anyone missing the filter properties? WTF? :eek: I take it you have not been hovering around video encoding forums such as this alot. I am not the first. Nor am I the last. Hopefully I will imagine all your time was tied up with improving WM9 over WM8's lack luster performance. ;) But if there is no use for such things then why do the Microsoft decoders themselves implement configurable decoder properties? What is the use in such things if one can not access them? Surely there must be a secret key combo or some string in the registry that can be altered to allow access in the new versions. If not perhaps there could be a patched or alterd binary to allow such. Because it is a reasonably big deal.

Originally posted by amirm
3. You say that On2 tried to keep up with WMV 8. WMV 9 outperforms V8 from 15 to 50%, depending on content. Having gone through the exercise of making WMV 9 better than V8, I have to say that it was a much harder job than outperforming less capable codecs such as MPEG-2/MPEG-4. So anyone who thinks they can make On2 better, has their work cut out for them.

You loose some credibility points there for me. DXN made such claims with Divx5 yet could not deliver. Microsoft has made such claims in the past without validating results to back them up. WM9 final will be tested. But claims of up to 50% improvement, heck even 15-20% are down right unbelievable. That is unless advanced aliens from another solar sytem are helping you with your algos. It sounds like so much hype. If all such claims were true I could losslessly compress a 3 hour DVD movie to a 250Mb CD in perfect quality

Originally posted by amirm
4. I see a lot of quality claims wrt to WMA and WMV but it is very hard for me to address them without seeing the clips, encoding parameters, tools used, etc., etc. One wrong parameter and you have apples and oranges comparison. The devil is always in the details.

If you stick around I am sure we can provide you with such clips for WM9 final. As for WM8, well since it did not have that many details to be configured, and the output was generally over all inferior to the output produced by inferior MPEG4 codecs I did not keep any clips on hand.

Originally posted by amirm
I implore everyone to do their own testing and not draw quick conclusions from casual tests. WMV has been compared and tested by experts alike and it always, and I mean always outperforms MPEG-4 derived codecs, not matter what their origin is (Divx,, etc.).

There will be testing to be sure. But one mans expert is another mans fool. Most in the past have tried to prove such things with slanted stats of which Microsoft is no stranger. Can you provide us with any links to 3rd party conducted, non Microsoft sponsored tests which can confirm this?

Originally posted by amirm
5. I keep seeing the comment that WMV is a "low-end" codec because we do well at streaming rates. But this is an improper generalization. WMV is scales very nicely with data rate and resolution because it is optimized differently for various "profiles" (another reason to not generalize its quality from encoding at one or two data rates). As a matter of fact, most of the kudos we receive are at the higher end of the spectrum. Here are some examples:

We have people projecting HD WMV on 30+ foot screens at data rates that are the same as standard resolution MPEG-2 on DVDs. See BMW Films Release (http://www.microsoft.com/presspass/press/2002/Oct02/10-30MotownPR.asp). And people can't tell the difference from film in surveys done after the viewing. There is tremendous amount of film grain in the "Shadows" movie and all of it is preserved like no other codec can. Sundance film festival just added WMV as one of its approved submission formats for independent films. No other compressed format is allowed (all the other formats are broadcast quality tape and 16/32-mm film).

I think I will deffer to temporance on that. I sense lots of room for skewed stats and the influence of pockets full of cash.

Originally posted by amirm
Here is a super long thread on avsforum on people encoding with beta version of WMV 9 their 1080p content. http://www.avsforum.com/avs-vb/showthread.php?threadid=171590&perpage=20&pagenumber=1

Here are some comments on the quality:

"Amazing. On my 2.8Ghz 1Gb machine it plays well Simply amazing quality"

"All I have to say is "Wow!" .. I'm looking very enthusiastically towards HDTV"

"<In the voice of Janice from Friends> Oh, my god! </In the voice of Janice from Friends>

That rocks! Now there's no turning back, either I have to emmigrate to the states or get me a HD pimp."

You get the picture I think :). BTW, this is where I ran into Nic.

Not everyone there is an over commited zealot but there are several. If you got some of them to convert I would be impressed. But far from convinced. Lets face it they are every where. We will just have to see for ourselves when the time comes. ;)

[Nic]

Too bad. Now that you mention Nic I think he was the one I had heard that from.
You never heard that from me! WM8 & WM9 (Oh and probably WM7 too) are supposedly very very different (although I (like most I assume) know very little about there exact inner workings)

However, I have stated that they're old codecs (MPEG-4 v1/2/3) are very similar to compliant MPEG-4...the ffmpeg source code (if you can navigate it) shows this clearer than I can explain.

Also, can we please not hold poor Amir accountable for all the "injustices" that some believe Microsoft have done. Im sure he's here just to enjoy the forum, he shouldn't have to justify the MS Corp. I'd hate to push a member away, because he knows everytime he posts he's going to get bombarded. (Im not saying that's what anyone in particular is doing BTW).

Cheers,
-Nic

[Rash]

Originally posted by trbarry
I would love to see a high bit rate codec shootout. ;)

- Tom

I encoded a hollywood studio logo using the best quality from DivX, XviD, WM9 and RV9. The quality is really impressive on all of them, but the smallest file is DivX.

DivX - 5.3Mbs
WM9 - 5.7Mbs
RV9 - 6Mbs.
XviD - 6.7Mbs

Sorry I can't provide you guys the clip because it wouldn't be legal. :)

[Atamido]

Originally posted by Rash
I encoded a hollywood studio logo using the best quality from DivX, XviD, WM9 and RV9. The quality is really impressive on all of them, but the smallest file is DivX.
Just because the quality is set to 100% on all codecs doesn't mean that they will all look the same. You will need to compare each of the clips to the original to determine which one looks best. While DivX produces the smallest file and XviD the largest, it may just be that XviD found ways to use more bytes to get a little bit more quality.

Please compare the quality of each and post back.

[trbarry]

Here is a super long thread on avsforum on people encoding with beta version of WMV 9 their 1080p content. http://www.avsforum.com/avs-vb/ show...20&pagenumber=1

Here are some comments on the quality: ...

The comments at AVS are true and the quality of the sample clip was impressive. You can still download it and see (recommended).

But you've got to remember that those of us who hang out on the AVS HTPC forum are mostly using home theater PC's with high rez larger screens to view DVD's at upscaled HD resolutions. So in addition to being about WM9 those comments also reflected:

1) A comparison of HDTV resolutions to DVD, which is quite favorable, and, maybe more importantly

2) Comments on seeing HDTV resolution video in a scenic Hawaaian setting of some very beautiful girls wearing only very skimpy bikinis. ;)

So now go read those comments again. ;)

- Tom

[Neo Neko]

Originally posted by Nic
Also, can we please not hold poor Amir accountable for all the "injustices" that some believe Microsoft have done. Im sure he's here just to enjoy the forum, he shouldn't have to justify the MS Corp. I'd hate to push a member away, because he knows everytime he posts he's going to get bombarded. (Im not saying that's what anyone in particular is doing BTW).

Cheers,
-Nic

No need to say. I know I am a pessimist when it comes to MS alot of times. But not without reason. And notice I am not giving major kudos to anyone at all in those recent areas. DXN has the most recent account to date of over hyped performance. I am leaving the judge out on WM9 for now. But I am getting skeptical. WM9 final will be given a fair shake soon. Rest assured it is all in good nature. Till then I'll keep nudging Amir for that libWM-9.x.x.i586.so stripped down decoder for the *NIXs. WM9 looks like it could be fun. But again a large group of people are being left out completely once more. :( BTW Nic I am getting ready to test yer new version tonight or tomorrow a bit. If only my linux media center could simply decode it to my tv. *nudge* *nudge* ;)

[Neo Neko]

LOL@TOM
Originally posted by trbarry
2) Comments on seeing HDTV resolution video in a scenic Hawaaian setting of some very beautiful girls wearing only very skimpy bikinis. ;)


Well at least they are not doing what DXN did with the porno. Or perhaps I should say to bad. :D ;)

[Rash]

Originally posted by Pamel
Just because the quality is set to 100% on all codecs doesn't mean that they will all look the same. You will need to compare each of the clips to the original to determine which one looks best. While DivX produces the smallest file and XviD the largest, it may just be that XviD found ways to use more bytes to get a little bit more quality.

Please compare the quality of each and post back.

In fact I did that and the quality on all videos are extremely close (no to say equal) to the original. Just some encoding artifacts around moving objects over a still background on DivX and XviD (but they are really really imperceptible, unless you really want to see them :)).

WM9 and RV9 are perfect to my eyes.

I know this might not be a great test, but this is the test I usually do to new codecs for myself. :)

[trbarry]

Well at least they are not doing what DXN did with the porno. Or perhaps I should say to bad.

Uhhhhhh, is there an important HDTV resolution Divx download sample that I missed? ;)

- Tom

[RadicalEd]

HD porn, lmao
Once I saw a womens diving competition on an HDTV in Circuit City and it was .. mm :O

[Psymaster]

Originally posted by RadicalEd
HD porn, lmao
Once I saw a womens diving competition on an HDTV in Circuit City and it was .. mm :O

The picture or the women?

[temporance]

Originally posted by RadicalEd
HD porn, lmao
Once I saw a womens diving competition on an HDTV in Circuit City and it was .. mm :O
What sort of diving?!?! :p

[amirm]

Nic,

Thanks again for the warm welcome to the forum. While I don’t mind have lengthy discussions in person, typing all of these replies is quite time consuming and while I could afford it during my holiday, I am about to go to work on Monday and will be flying out to Las Vegas for the Consumer Electronics Conference to boot. So I won’t have the opportunity to spend as much time here. That’s why I was trying to focus the discussions around specific questions or actual benchmark tests.

Quickly answering a few of the questions posed:

The BMW films project was encoded at 8 Mbit/sec, using constant bit rate (CBR). Yes, we would have preferred VBR encoding for these but another post house did the work and for some reason decided to use CBR. As you can imagine, using 8 Mbit/CBR puts us at a serious disadvantage to DVD’s VBR encoding but we sill managed to create great results and preserved the aforementioned film grain. We were encoding 2.2 times more pixels than DVD with lower (maximum) data rate and hard ceiling on bit allocation! BTW, the sound track was WMA Pro 7.1 so we had more audio channels than DVD.

Comment was made that 16 mm film grain may be easy to encode. I don’t see how the correlation in film grain is going to improve from frame to frame just because the grain is larger in 16mm. But even if this is true, the grain still requires additional bits to encode.

Imagine a black area which in film domain is full of colorful dots. This detail will require a lot more bits to encode than the solid black area shot using HDcam video camera. Still, this is not the point I was making. My point was that competing codecs with strong in-loop filters (Real, H.264) completely filter out such detail. Other simpler codecs like MPEG-2/MPEG-4 attempt to encode the grain, run out of bits, and you get distortion that looks nothing like the original. We manage to get good efficiency to encode the film grain without resorting to strong filters. You can always add a pre-filter to our codec if you want to get rid of the source noise but can’t create detail that is already lost (i.e. can’t do the inverse).

Neo asked for independent tests showing we are better than MPEG-4. Here is one reference to tests done by PC Mag by Jan Ozer who is a long time reviewer of codec technology, author of a number of books on multimedia and video compression who btw, is traditionally very hard on Microsoft: http://www.pcmag.com/article2/0,4149,704,00.asp


Here is one quote from his article: "The second, more critical issue is that MPEG-4 video quality has consistently ranked behind that of Microsoft's Windows Media Technologies".

As with any other test, you could argue with how the testing was done, who did it, what tools they used, etc., etc. and hence the reason I don’t reference such things often. But you (fairly) asked for more data and here is one piece. Take it for what it is worth. Note that this comparison was against WMV8 and not WMV 9.

Let me finish by saying that you all have access to our technology. My only request which I repeat again is to use it long enough to get your own opinion of it. You don’t have to fork out a penny to get this knowledge. As a minimum, you will have some hard data to beat me up with next time :).

Thanks,
Amir

[temporance]

Originally posted by amirm
Thanks again for the warm welcome to the forum. While I don’t mind have lengthy discussions in person, typing all of these replies is quite time consuming....
Your time is really appreciated, thanks Amir.

My point was that competing codecs with strong in-loop filters (Real, H.264) completely filter out such detail. Other simpler codecs like MPEG-2MPEG-4 attempt to encode the grain, run out of bits, and you get distortion that looks nothing like the original. We manage to get good efficiency to encode the film grain without resorting to strong filters.
These are interesting points. It seems that the industry is starting to recognise that different techniques are better at different bitrates, for example loop filtering is better for lower bitrates but not useful (and even undesirable) at higher rates.

So, your codec has no loop filtering, ergo it is probably not as good as Real or H.264 at lower bitrates. But this doesn't matter since on these forums we're interested in higher bitrates. The absence of loop filtering also backs up your claims for a more lightweight decoder.

I'd love to have an open, technical discussion about WMV9. Let's make an analogy with cars. All are car-lovers on this forum. Most just like to test-drive the cars and see how they perform. Some like to look under the hood and discusss whether we want ABS, v8, rear-wheel-drive, etc (that's me). Some even want to know if they can soup up the car by tweaking the engine management or altering the fuel. And most of us want to be able to drive our chosen car on non-Microsoft roads without invalidating the warranty. ;)

Now Microsoft's much-hyped new car is due off the production line and a few of us have already taken the prototype for a spin. Others of us are frustrated at the sealed hood and censored pages in the spec. We're guessing and trying to learn things by deduction and would love it if you told us why WMV9 should be better than standards-compliant codecs. Having said that, I'm sure we're all looking forward to a good long drive of the final release.

[karl_lillevold]

Originally posted by temporance
These are interesting points. It seems that the industry is starting to recognise that different techniques are better at different bitrates, for example loop filtering is better for lower bitrates but not useful (and even undesirable) at higher rates.


It is interesting to note that it seems Microsoft is alone against the rest of the world on whether in-loop filtering is useful or not. H.264, jointly developed by the ITU and MPEG, has in-loop filtering. If you look at the list of participants in ITU and MPEG, you will find that it is almost everyone working or having an interest in video compression. Repeated experiments by the participants have found that a high quality, adaptive in-loop filter, significantly improves compression performance (since the encoder does not have to constantly re-encode edges and ringing).

In H.264, the in-loop filtering is highly adaptive, variable, and tuned such that it does not hurt performance at high bitrates, and it is even customizable by the encoder. So in this respect, I disagree with Amir, who says "My point was that competing codecs with strong in-loop filters (Real, H.264) completely filter out such detail."

However, the in-loop filter in H.264 is extremely compute intensive, probably around the same as a full decode, because it needs to adapt to image content. It requires extensive optimizations to not completely sink a decoder. I think this is a very likely reason why a codec might choose not to use an in-loop filter with the adaptivity needed.

Originally posted by temporance
So, your codec has no loop filtering, ergo it is probably not as good as Real or H.264 at lower bitrates. But this doesn't matter since on these forums we're interested in higher bitrates. The absence of loop filtering also backs up your claims for a more lightweight decoder.

Now over to RV9. It sounds like many in this forum have found that it can stand on its own and more against the competition, at the high bitrates most oftentimes used here. And even though it is slower than previous generation codecs, similarly to WMV9, HDTV 720p and 1080p decode is possible on P4 2.4+ GHz systems, especially with hyper-threading or multi-CPU systems, since the core decoder is fully multi-threaded (which at least WMV9 Beta did not appear to be).

and a little addition:
Originally posted by amirm
Here is one quote from his article: "The second, more critical issue is that MPEG-4 video quality has consistently ranked behind that of Microsoft's Windows Media Technologies".
I like the full sentence better :)
"The second, more critical issue is that MPEG-4 video quality has consistently ranked behind that of Microsoft's Windows Media Technologies, RealNetworks' RealVideo 9, and Sorenson Media's Sorenson Video 3."

[amirm]

Regarding in-loop filters, a conclusion was drawn that we do not use in-loop filters. But if you read my post, you see that I never disclosed what we do or do not do (wrt to in-loop filters or otherwise :)). I simply said that technologies that opt for strong in-loop filters, tend to demolish detail that is noticeable at high resolution and quality levels. This is definitely true of H.264. Anyone who thinks otherwise, has not done enough testing of H.264 (as the chair of the H.264 standards group, I assure you that we keep an eye on the qualities of this codec :-).

Karl says that in-loop filters are accepted as a good technique in compression circles. I partially agree with this. The benefits of in-loop filters are definitely there at lower bitrates *and* resolution. But recent HD testing (subjective and objective) shows that at least in the case of H.264, the strong filter is damaging to the picture detail. Film grain is lost and so is resolution. Using temporance’s car analogy, in-loop filters seem to be like soft shocks. They give great ride at low speed but high-speed cornering seems to suffer.

Note that RV9's filtering appears even stronger than H.264 (or less adaptive). It also tends to be unstable at times, resulting in picture detail that comes and goes (with motion) which can be very annoying when you are looking at a large HD images on a projector/big-screen HD set (same thing is happening at lower resolutions but is harder to spot). So while RV9 may allow the user to encode at high resolutions, I do not consider the output to be considered "HD" if "HD" is defined as picture that always has better picture detail than SD resolution image. MPEG-2 at SD resolutions is preferable to RV9 “HD” sometimes because of its stable output. Reading the reports on RV9 at SD resolutions in this forum, seems to bear out the same conclusion as evidenced by the number of requests to "turn off the post filter" in RV9.

As to temporance's comment that lack of in-loop filter means we are not as good in lower bit rates, this is again not correct. As I mentioned in other posts, WMV has a number of algorithms that kick in at various bit rates and resolutions. Indeed, we are quite a bit more aggressive at lower bit rates in trading off resolution/detail for typical compression artifacts. I will not disclose how we do this (i.e. whether it is due to in-loop filter or something else). But I will say that our performance gap with MPEG-4 is even higher at lower bit rates.

Related to above, at 320x240 resolution and lower, we have a frame interpolator which uses optical flow analysis (i.e. tracks every pixel individually to figure out where it would have been in missing frames). This allows us to encode at say, 10 fps, and then playback at 30 fps. This gives us another unique advantage since it takes less bits to represent 10 fps versus 30. If you take this into account, our performance advantage becomes even higher at lower resolutions.

As to CPU usage, while Karl is hoping for faster machines to play their HD clips, I am typing this on a Dell laptop which comfortably decodes 720p WMV9 with 6 channel WMA Pro audio! Needless to say, any desktop machine bought in the last two years comfortably handles WMV9 720p. We also play 1080p on 3 Ghz P4 (2.8 Ghz can also decode most clips). To put this in context, we are decoding 24, 2-Megapixel images (1920x1080) per second and sending them to the GPU – this is a ton of data to handle and move to the frame buffer!

Amir
Microsoft

[oddball]

There is one HUGE probelm with WMV9 (and RMV9). Lack of support for SPDIF AC3 (Or DTS) output. It's one area in which MPEG4 variants gets my vote. I don't use a 5.1 PC speaker set. I output using AC3 passthru to my AC3/DTS reciever and let that send to some good 5.1 hi-fi speakers. WMV9/RMV9 etc *ALL* do surround sound by sending each channel to the outputs of a 5.1/6.1 audio card. Not SPDIF passthru (As this would require the signal to be encoded to AC3/DTS realtime. Something that is not quite there yet. Although I do believe someone on this forum is working on such a scheme. All I can say to that though is that it would put additional CPU stress on the PC. So bang goes any chance of playing such files on a P3 700 without bringin the system to it's knee's).

I will always prefer MPEG4 for this very reason.

Please correct me if I am wrong on this one.

[The Belgain]

Needless to say, any desktop machine bought in the last two years comfortably handles WMV9 720p.

Not quite: I have a 1100 MHz Athlon (with DDR memory) which was pretty much top of the range when I bought it exactly 2 years ago, and it can't quite decode the 720p WM9 bikini clip realtime (I get about 20 fps decoding, but the audio stutters horribly and the video occasionally stops completely for a second or so before going on). Most people buying a desktop 2 years ago would have been going for an 800 Duron or Celeron, so that claim doesn't quite hold.

Also, why do I not get "smoothish" playback? With an XviD clip at that res, I get slightly slow playback, but the audio plays fine and the video stays in synch (just drops a few frames per second)? I tried in both ZoomPlayer and WMP9 (doesn't work at all in WMP6.4).

[RadicalEd]

Originally posted by Psymaster
The picture or the women?

both ;O

Originally posted by amirm
Needless to say, any desktop machine bought in the last two years comfortably handles WMV9 720p.

just like Belgian said, I could barely playback the Bikini clip on this 1ghz Athlon w/ 384 sdram that I bought a year and a half ago :\

[amirm]

I stand corrected. Perhaps "any" was too strong of a word :). WMV9, 720P requires P4 2.2 Ghz equiv. performance. I agree not everyone has been buying such machines for the last two years.

On the issue of S/PDIF, this is kind of correct at the moment. There are some interim solutions though. Nvidia AMD chipset does have an on-board hardware AC-3 encoder (similar to what is in Xbox btw) which converts the multichannel system sound (i.e. WMA Pro) into Dolby AC-3 on the fly. So there is no CPU hit for this.

If you don't have an AC-4 encoder in your system but set your system audio config to stereo, WMA Pro on the fly converts its multi-channel output to (stereo) Matrix surround (through both S/PDIF and analog outputs). So any external reciever with Dolby Pro Logic should get a surround experience, but of course, it will not be as good as straight WMA Pro.

Note that MPEG-4 does not bring any solution by itself to the issue of S/PDIF. As a matter of fact, MPEG-4 has nothing to do with Dolby and using AC-3 creates a non-compliant MPEG-4 stream. What you are saying is that the tools allow you to create non-MPEG-4 streams by embedding AC-3 in AVI or similar container together with MPEG-4 video. Same approach can be used for WMV 9 although the tools are not quite there to do this just this minute.

Finally, AC-3 at 384kbps uses up a lot of your storage space (~30% at 1.3 Mbit/sec total rate). You can encode in WMA Pro at half the data rate and use the savings to make the video look better. Sure, for now, you would have to listen to it in Matrix surround but when you upgrade your sound card, you get your discreet output back and still have the bit savings.

Amir
Microsoft

[karl_lillevold]

Originally posted by amirm
Regarding in-loop filters, a conclusion was drawn that we do not use in-loop filters. But if you read my post, you see that I never disclosed what we do or do not do (wrt to in-loop filters or otherwise :)). I simply said that technologies that opt for strong in-loop filters, tend to demolish detail that is noticeable at high resolution and quality levels. This is definitely true of H.264. Anyone who thinks otherwise, has not done enough testing of H.264 (as the chair of the H.264 standards group, I assure you that we keep an eye on the qualities of this codec :-).

But recent HD testing (subjective and objective) shows that at least in the case of H.264, the strong filter is damaging to the picture detail. Film grain is lost and so is resolution.

Hmm, this is interesting. I always thought of the development of H.264 as a collaborative effort, where participants (generally) try to contribute such that the result turns out to be a better standard. Based on what you say here, if true, it seems Microsoft has held back information that would have made H.264 even better than it already is.

[amirm]

Karl, the accusation is disingenuous and makes assumptions not in my post.

As you well know, H.264 is on a march to finish the standard, not to re-write the algorithm with every new piece of data. The goals of H.264 (good quality at any computational price) have been met and the organizers are anxious to finish the standard.

Now, if you are telling me that it is news to H.264 experts that in-loop filters reduce video resolution in some cases, then I would say you don’t know the same experts we do :). After all, if users in this forum can detect such a feature from casual testing (as in softness in your codec), experts in the field would know it in a heartbeat.

Regardless, the information that I talked about has been sent to JVT as a "liason" report by the people who ran the tests (Microsoft did not conduct the tests or owns the results). So nothing has been kept back.

Amir
Microsoft

[temporance]

Originally posted by amirm
As to temporance's comment that lack of in-loop filter means we are not as good in lower bit rates, this is again not correct. As I mentioned in other posts, WMV has a number of algorithms that kick in at various bit rates and resolutions. Indeed, we are quite a bit more aggressive at lower bit rates in trading off resolution/detail for typical compression artifacts. I will not disclose how we do this (i.e. whether it is due to in-loop filter or something else).
Reading between the lines here, and picking up on other comments and observations I've made, I think this means that you do use an in-loop filter at low bitrates. In other words, yours is simplest form of an adaptive in-loop filter which is turned on/off according to bitrate/resolution. RV9 and H.264 have much more sophisticated adaptive loop filtering and their filters are also likely to be turned off completely at high bitrates. OK, so they may need more tuning but effectively we all have the same technology.

On film-grain, we all agree that, at high bitrates, it would be desireable not to filter out the film grain on the output of the decoder. But it would be a good idea to remove film grain by filtering in-loop as the grain is not correlated between frames and not useful for motion compensation. So one could envisage a decoder where two different post filters are used: one for the decoder output and one for the MC loop. Perhaps WMV9 does this, no-one will ever know. ;)

But I will say that our performance gap with MPEG-4 is even higher at lower bit rates.
Aha! So the quoted (and much hyped) performance figures would be based on low-bitrate performance!! Implying that for DVD rips, WMV9's performance is similar to DivX (MPEG-4).

Related to above, at 320x240 resolution and lower, we have a frame interpolator which uses optical flow analysis (i.e. tracks every pixel individually to figure out where it would have been in missing frames). This allows us to encode at say, 10 fps, and then playback at 30 fps. This gives us another unique advantage since it takes less bits to represent 10 fps versus 30.
This is the sort of trick that appeals to marketing types, or people who don't have much experience of video, and is typical of the ideas proposed by newbies on these forums. Given, it could reduce jerkiness by providing some consistent motion between frames. At best though, it makes the video look even more processed and artificial, and at worse low-bitrate video coding artifacts cause the interpolation to do very strange things. If it does work well, we are sure to see people applying the technique to the output from other decoders - so it's not really a pro-WMV9 argument. It's like telling me your car comes with optional seat covers when I still don't know what engine it has. Having said that, congratulations on getting it working and I am sure to have a play with the new toys!!!

[-h]

On film-grain, we all agree that, at high bitrates, it would be desireable not to filter out the film grain on the output of the decoder. But it would be a good idea to remove film grain by filtering in-loop as the grain is not correlated between frames and not useful for motion compensation.

Sure, PNS for video instead of audio. However this would harm the PSNR values that video people in the field obsess over (often to a fault in my opinion).

This is the sort of trick that appeals to marketing types, or people who don't have much experience of video, and is typical of the ideas proposed by newbies on these forums. Given, it could reduce jerkiness by providing some consistent motion between frames.

Temporal interpolation is an old trick, but would be very effective for low-bitrate encodings (actually it would help at any bitrate if done well-enough, format support through side information would be a great aid). I know Alexis Tourapis published a couple papers on low-cost temporal interpolation, as well as being a recognised expert on optimising block-based motion compensating video encoders. The fact that he's been working for Microsoft as a visiting researcher for some time now gives me a good feeling about the internals of WMV9, much better than WMV7/8.

-h

[amirm]

Temporance,

I admire your desire to gather information by “reading between the lines.” But as before, the information in my posts is not sufficient for you to arrive at conclusions that you are stating. I repeat again that I never said we do or do not use in-loop filter or any other techniques. We may know about algorithms that others do not that work better than in-loop fitlers. But then again, may be not. The only thing you know for certain is that we haven’t disclosed what we have :).

Aha! So the quoted (and much hyped) performance figures would be based on low-bitrate performance!! Implying that for DVD rips, WMV9's performance is similar to DivX (MPEG-4).

So let me understand this better. You are agreeing that we have a codec that is twice as good as MPEG-4 at say, 300Kbps. But that once the data rate gets to 1 Mbit/sec and resolution climbs to 720x480, the quality difference becomes zero??? I know you are more logical than this :).

The day our codec looks as bad as Divx “at DVD rip” rates, is the day I change my name and learn to walk backwards :). I am that confident of quality of our codec, regardless of the data rate and resolution you pick.


"This [frame interpolation] is the sort of trick that appeals to marketing types, or people who don't have much experience of video, and is typical of the ideas proposed by newbies on these forums… At best though, it makes the video look even more processed and artificial"


How do you think a $100,000 Snell and Wilcox standards converter works? When you watch BBC programming on PBS, do you think it looks "processed and artificial"? Upsampling from (BBC’s) PAL 50 Hz to NTSC at 60 Hz requires an advanced interpolator like the one that we implemented. Years of research have gone into "proper" upsampling techniques. Calling them "marketing" features ignores the reality of the science and technology already available.

Sure, there are cheap “upsamplers” that use simple cross fades between frames and produce the results that you talk about. But that’s not what we have.

Now, I am not going to tell you that the technique works perfectly all the time (we do have limited MIPS to work with). But the experience is far better than the picture you are painting.

Amir
Microsoft

[temporance]

So let me understand this better. You are agreeing that we have a codec that is twice as good as MPEG-4 at say, 300Kbps. But that once the data rate gets to 1 Mbit/sec and resolution climbs to 720x480, the quality difference becomes zero??? I know you are more logical than this :).
No, I just know that if you are going to quote a 50% improvement then you are going to base that maximum on the best possible scenario. Your [marketing department's] claims have to be taken with a pinch of salt for now :)

How do you think a $100,000 Snell and Wilcox standards converter works?
I know how S&W products work. They use costly motion estimation (phase correlation - although the algo isn't so important) and advanced interpolation schemes implemented in FPGA's, hence the expense (and heat dissapation). It is because I know how hard it is to do this job well that I am sceptical about the performace that can be achieved in sub-real-time on a PC and with a less-than-perfect source. Still, this is low-bitrate video we're talking about, not broadcast-quality.

[bond]

I dont want to make promotion for m$ but at their homepage (http://microsoft.com/presspass/press/2003/Jan03/01-079SeriesFinalReleasePR.asp) (Series9 Final Release) they write:

The terms enable licensees to develop implementations of Windows Media Audio and Video 9 Series codecs in a wide range of environments ... and to include the Windows Media codecs in their products in any file container and at costs significantly lower than for other technologies ... There is also a significant content use fee for MPEG-4, while there are no content use fees for the Windows Media Audio and Video codecs ...

... By enabling the use of Windows Media codecs independently of ASF, software vendors that support encoding, editing or playing back content in containers such as AVI or MPEG can quickly and easily include support for the superior compression of Windows Media Audio and Video 9 Series in their applications.wm9 content in avi, ogm... sounds interesting :)

[RadicalEd]

I too was wondering about that. I came to the conclusion that I must be misreading and it probably was to be taken to mean something like Premiere's WMV export plugin could be created for programs that can also handle mpeg and avi. Not sure though, I'll be waiting eagerly to hear amir's clarification of it :\

[Rash]

So, is it true that the "final" version of the Encoder and the Player is out?

Yes pal. It would be really amazing if we could encode WM9 from VirtualDub or Adobe Premiere. :)

[alan_erickson@hotmail.com]

I like the quality of the wmv codec. Unfortunately microsoft doesn't want you to be able to encode it to anything else, as evidenced by the message I get when I try and import a wmv into VirtualDub. That message being:

"Not supported Microsoft owns US patent #6,041,345 on the ASF file format, preventing third party applications from gyrating data from ASF files ASF file format support was removed as of V 1.3d at the request of Microsoft and to avoid patent infringement claims, and as such VirtualDub no longer supports ASF Please do not ask for versions that do."

amirm may come in here spewing a lot of positive words, but the actions of microsoft speak quite differently.

[Rash]

Yup, I kinda knew that, unfortunately. :(

[pcdvdguy]

This question is kind of aimed at amirm...

In the WM9 product features section, Microsoft claims :
...
# Hardware-based Windows Media Video Acceleration
Experience improved performance throughout with the next generation of video cards that support DirectX® Video Acceleration (DxVA) technology. With these cards, the rendering of Windows Media Videobased content can be offloaded onto the video card processor, making video playback smoother and making video playback at higher resolutions possible.
...

Does any current (or near-future) video-card *actually* support WM9 acceleration? The reason I ask, is that no MPEG-4 codec (Divx, Microsoft, Xvid) currently uses DXVA-acceleration, even though that would be a great way to reduce CPU-usage. [On a different forum, someone from ATI explained that the (original) ATI Radeon did not support all the motion-compensation modes of MPEG-4. Does the Radeon9500/9700 support them?!?] Furthermore, neither Nvidia or ATI claim to accelerate MPEG-4 playback, only DVD/MPEG-2.

The other thing I noticed, is that the DirectX VA API specification has restricted profiles for H.261, MPEG-1, MPEG-2, and H.263. There is no specific profile for MPEG-4, although the API-document implies MPEG-4's the iDCT/motion-comp operation are derived from H.263. (I'm not familiar with the details behind MPEG-4 and H.263, so please correct me if this is a false statement.)

[amirm]

There are currently no GPUs that accelerate WMV although our partners are working on it. A gating item was releasing WMV 9 that we just did. I will report when these solutions come to market.

Amir
Microsoft

[amirm]

Originally posted by ChristianHJW
... if hell freezes over maybe ;) ...

This comment was made wrt to availability of VCM/VfW codec for WMV. I am pleased to report that it was a really cold day in hell on January 7th when we announced that we will indeed, be releasing such a component :). The tentative date is next month.

Another question was asked wrt to having to use our file format and audio format with WMV. We announced that you can now mix and match our components at will. With the above VCM, you can for example create AVIs with AC-3 audio streams (which addresses the other concern someone else had).

Amir
Microsoft

[ChristianHJW]

Originally posted by amirm This comment was made wrt to availability of VCM/VfW codec for WMV. I am pleased to report that it was a really cold day in hell on January 7th when we announced that we will indeed, be releasing such a component :). The tentative date is next month.

I am impressed ! Not more to say right now, but i have to admit i am really impressed by this decision from M$. Seems as if they were pretty successful in promoting their codec to hardware companies, so they can see things easier now with respect to software and how the codecs will be used. In any case, i appreciate that. Competition is the salt in the soup of Life :D ...

Another question was asked wrt to having to use our file format and audio format with WMV. We announced that you can now mix and match our components at will. With the above VCM, you can for example create AVIs with AC-3 audio streams (which addresses the other concern someone else had). Amir

Again, a very interesting decision. Does this mean M$ was paying the Dolby licensing fees for us ? It will be very interesting to see if these ACM codecs will be compatible with existing AC3-AVIs created by Nandub, VirtualdubMod or AVImux-Gui.

[alan_erickson@hotmail.com]

Amir, will Highmat support wmv8, or just 9?

[midiguy]

yesh lee zayin gadol meod :D :p

[lyl797]

WMV9 vs DivX 5.x

Video quality ---- Same to me, tiny difference is way beyond normal human eyes' recognition.

Internet Streaming Video
WMV9 --- You can do streaming Video with WMV9, and the play back thru internet is very smooth with WMP9,
you can put in several different bit-rate coding into one WMV file that will fit into
different connection speeds.

In another word, I can make a WMV which can be stored on CD, DVD; and the same file could be used
on internet streaming video broadcast.

Divx --- I still don't know how to make streaming Video out of Divx ?


Encoding
WMV9 --- User friendly native encoding program from Microsoft,

DivX --- Need 3rd party Encoding programs to smooth the operation.
and there are lots of them in the market, sometimes, it's really hard
to keep track of all the updates for them and not easy to decide which one
to use.

Player
WMV9 --- Comes with windows OS free, update thru Windows online update.
No need to keep track of the updates ...

Divx --- Need to download, install, update .......

[MaTTeR]

Why compare to DivX 5.x? IMHO if you really want to see how WMV9 compares to the best that MPEG4 has to offer then you must compare it to XviD.

No offense meant but stating that it's far superior than the results of DivX 5 isn't really saying much for WMV9 IMO. If the same were said about XviD then I might be bothered to download the software and try it out myself;)

[oddball]

Plus you don't get to use all those nice filters in WMV9. I'd like to see how it handles anime interlacing and subtitles etc ;)