AVS Forum banner
Status
Not open for further replies.
1 - 20 of 53 Posts

·
Registered
Joined
·
9,884 Posts
Steve -


For HDTV? Do you really mean that low a bit rate?


As a very rough guess I usually figure it will take a minimum of 1/4 bits per total pixel tranmitted for decent HDTV quality.


So for, say, 1280x720 @ 24 FPS it would need at least 5.53 mbps.


Some brief further 1 pass CBR tests this afternoon suggest that it's similar for both WM9 and Xvid. And I'm running some 2 pass VBR tests now.


- Tom
 

·
Registered
Joined
·
5,944 Posts
Discussion Starter · #3 ·
Maybe I am reading the encoder wrong


In teh VCM wm9 screen what bit rate are you piutting in?


for 720p material I have been entering

3,000,000

\\

This is in bits - correct???



I am using UNDOT() and light deringing to remove some MPEG2 artifacts




With 2-pass encoding I get about 1GB/42minutes (5:1 compression) and really think it looks transparent.
 

·
Registered
Joined
·
9,884 Posts
I was just using my experience with Xvid. I don't have a feel for WM9 yet except for the one pass CBR tests where it gives about the same results as Xvid, at least measured by PSNR for the same size files.


But I was using about twice as many bits. I haven't had a chance yet to see how gracefully WM9 degrades as you start starving it so maybe I'll find it would be completely happy with half the bit rate. Dunno yet.


Right now I'm doing multiple encodings of some deinterlaced Masters Golf, resized to 1440x800 @ 30 fps.


When did Zoom Player stop reporting the actual play frame rate? Is that just an XP thing? I always used to rely on that but it doesn't show it now.


- Tom
 

·
Registered
Joined
·
5,944 Posts
Discussion Starter · #5 ·
I have been using MPLAYER2 :) I find Zoomplayer amazing but confusing as hell :) Ditto with FDSHOW


In some respects I feel like I need an idiots guide to FDSHOW and Zoomplayer.


I am getting the true FPS at 3mbit 720p material 23-24 FS on Mediaplayer2


I tried going highter on the bit rate. Did some comparisoons with some very picky friends an good computer monitors and it was really tough t see the difference in still captures of Alias.


Skin tones, Jennifer Garner details etc were fantastic.


In fact, even at 2mbit most was really good (2-pass now) However, there are some scenes in one episode with rain that KILLED teh encoder and looked like **** at 2mbit. I will cut the AVI tonight and place it on my FTP site for you. It is a great into :)


Steve


Oh, again at 960x540 (then cropped) 800kbit looks fantastic VBR. In fact, the encoder is not even using it all it seems as I am getting about 300megs/42 minutes.


My friend commented - "you made the episode look better" - not sure betterbut the MPEG2 upconver artifacts are largery gone. :) thanks to


UNDOT()

MPEG2DEC3 CPU=5


I am really really impressed with CPU=5 then making it a bit lighter. The dereinging and deblocking is REALLY fast compared to some other filters.


The set top boxes are coming everyone .....

Sh....................


Don't encode PEAKS over 12mbit ....


Steve
 

·
Registered
Joined
·
9,884 Posts
You could be right here and I've gotten too picky and am using too high a bit rate. On my easy to encode clean golf material the Xvid came out undersized, usually a symptom that it doesn't really need it all. But I've come up with my current habits from compressing movies, which have the stupid film grain to deal with, both in my own encodings and in the artifacts they produce by stressing the original MPEG-2 encodings.


I'm repeating with a slightly lower bit rate.


- Tom
 

·
Registered
Joined
·
9,884 Posts
but I cannot confirm that WM9 compresses any better than Xvid in a high bit rate HDTV comparison.


I tried it a few ways but I'll post the results from 2 of them here. All my tests today were on a few minutes of CBS Masters Golf, deinterlaced and resized to 1440x800.


I used high enough bit rates that all results were indistinguishable from the (resized) original, so that part was a wash.


But I also measured PSNR using the free PSNR4AVI from Vanguard Software . That page has the download and also a good explanation of what PSNR means. There was one quirk in the test though. It seems PSNR4AVI always stopped after 2981 records for some reason, so PSNR was really tested for only about the first half of my output files. This should probably be okay as the same thing happened for both the WM9 and Xvid files.


PSNR2AVI displays the PSNR separately for the Y,U,V planes and then a combined average for all of them.


The first test I made was for a single pass constant bit rate (CBR) at a high rate of 8.64 mbps. I've been recently doing more CBR encodings because I'm more concerned about the peak bit rates for playability and CPU considerations than I am in saving the last nickel of disk space. But for these fairly clean easy-to-compress golf clips this rate was maybe a bit of overkill. Anyway:


WM9 One Pass CBR Results:
Quote:
WM9 - One pass CBR at 8.64 mbps

File size 220,682 KB


Average PSNR (Y U V): 42.10, 46.11, 46.84

Total Average PSNR: 45.02
Xvid One Pass CBR Results:
Quote:
Xvid - One pass CBR at 8.64 mbps

File size 220,792 KB


Average PSNR (Y U V): 42.37, 46.47, 47.99

Total Average PSNR: 45.61
The second test was a two pass VBR using both WM9 and Xvid at 6.5 mbps. I also set a peak bit rate on both of them at 8 mbps, again for CPU playability.


Wm9 - Two pass Peak VBR results:
Quote:
WM9 - Two pass Peak VBR at 6.5 (8 max) mbps

File size 169,756 KB


Average PSNR (Y U V): 42.07, 45.86, 46.72

Total Average PSNR: 44.88
Xvid - Two pass VBR results:
Quote:
Xvid - Two pass VBR at 6.5 (8 max) mbps

File size 163,960 KB


Average PSNR (Y U V): 41.93, 46.12, 47.67

Total Average PSNR: 45.24
As you can see, Xvid produced a slightly better PSNR and the same or smaller file size in both tests. So, at least for high bit rate HDTV compression, I really have not been able to confirm the claims that WM9 is twice as efficient as MPEG4. I think it may somewhat depend upon the codec implementation tested.


I posted pictures of the original (resized) and the two compressed VBR clips at:

www.trbarry.com/CBS_Masters_Original.jpg (huffyuv lossless)
www.trbarry.com/CBS_Masters_WM9_VBR_6.5_mbps.jpg WM9 , and
http://www.trbarry.com/CBS_Masters_X...R_6.5_mbps.jpg Xvid


I still don't really know what I'm doing with WM9, so if anybody has any suggestions how I really should have done it (buffer sizes, etc.) pls let me know and maybe I'll try it again.


- Tom


edit: see also below, where I added one more pair of tests at a suggested lower bit rate
 

·
Registered
Joined
·
1,750 Posts
Tom,


That's still a pretty high bitrate for WM9. For re-encoding, I usually shoot for ~.4X the MPEG2 rate. If the source was 1920x1080 video at 16 Mb/sec, that would be ~3.5 Mb/sec VBR WM9 for 1440x800. It would be interesting to see the relative numbers at that level of compression.


Joe
 

·
Registered
Joined
·
9,884 Posts
Joe -


I'll probably try one more pair of Peak VBR tests at a lower bit rate. But first I'd like to solicit some more opinions on things like performance, buffer sizes etc.


BTW, I was not using ConvertToYUY2 so I think I was passing WM9 straight YV12, from Avisynth 2.5. It seems I remember a problem someone stated with that but it didn't seem to be a problem for me. Anyone remember that problem?


- Tom


edit: and everybody keep in mind this is video material at 30 fps so we maybe have to adjust the bit rate up a bit. OTOH, that is probably more than compensated by the lack of film grain.
 

·
Registered
Joined
·
1,750 Posts
The performance slider 1 click from the left gives the best speed with maybe 5 - 15% higher file size compared to sliding all the way to the right - which is sloooooow. I haven't found any difference in quality or size with different buffer settings.


No problem with YV12 here.


Joe
 

·
Registered
Joined
·
9,884 Posts
Okay. I'll make up something for the buffer settings and run another test at 3.5 with a peak of 7. I'm running the Xvid version of it now.


- Tom
 

·
Registered
Joined
·
9,884 Posts
I did one more final pair of tests using the lower bit rates suggested above.


Wm9 - Two pass Peak VBR results:
Quote:
WM9 - Two pass Peak VBR at 3.5 (7 max) mbps

File size 97,984 KB


Average PSNR (Y U V): 40.04, 44.98, 45.97

Total Average PSNR: 43.66


Xvid - Two pass VBR results:
Quote:
Xvid - Two pass VBR at 3.5 (7 max) mbps

File size 98,218 KB


Average PSNR (Y U V): 40.65, 45.17, 46.82

Total Average PSNR: 44.21
It looks like that bit rate was quite sufficient for this clean low motion clip on both codecs. Xvid still had a slightly better PSNR value. Pics of the two new clips are at:

www.trbarry.com/CBS_Masters_WM9_VBR_3.5_mbps.jpg WM9 , and
http://www.trbarry.com/CBS_Masters_X...R_3.5_mbps.jpg Xvid


I forgot to mention it above but all the Xvid clips appear to play relatively smoothly, using Zoom Player on my 2.4 Ghz P4. But with WM9 there is some jerkiness on pans and zooms, suggesting dropped frames. This is true to a lesser extent even with the 3.5 mbps clip.


Note that in Xvid to get better playability I am encoding without any of the newer fancier and slower options like b-frames, global motion compensation, or quarter pel.


- Tom
 

·
Registered
Joined
·
1,750 Posts
Tom,


For that clip, you still haven't reached the sweet spot for WM9. Keep dropping the bitrate - 3, 2.5, 2, etc. until there is a significant drop in PSNR.


Dropped frames with a 2.4G CPU is puzzling. People are reporting 0 dropped frames from 1920x1080p24 using the 3G P4. Are you using 100% CPU during playback?


Joe
 

·
Registered
Joined
·
9,884 Posts
Joe -


Even using the 3.5 mbps Xvid I show over 80% CPU so maybe it's a driver or installation issue. (all WM9 clips were 100) I'm running Zoom Player with ac3filter on a newly installed XP system that I only put up to play with WM9. So maybe I need Radeon drivers or a DirectX update or something.


I haven't decided whether to keep it this way but I'll tune it if I decide to stay with XP on this one machine. For now I don't think I'll worry about the performance issues.


- Tom
 

·
Registered
Joined
·
5,944 Posts
Discussion Starter · #16 ·
Is XVID really that much better then DIVX5.03? I found at

720p (1GB/43 minutes) Divx looked horible.


I agree - the picutres look great.


Would adding the fancy XVID features improve it even more?


Maybe we should all be using XVID ....
 

·
Registered
Joined
·
5,944 Posts
Discussion Starter · #17 ·
Well,


MS is going to give my some stuff that trashes the XVID codec. The "green" is the masters is really the worst case senario for them. The agree that live 30FPS "green" picutre will be close if not any better. However, we will see what video they give me to try. Again, howevr, they say that at best it will be = or better. .... the data will tell...


Are you likeing XP?? I don't know what is wrong. With the newest Radeon drivers and VMR I can play 1080i at 60% CPU usage TS streams :)



I will get these clips next week - we need a good FTP site to share them


Steve
 

·
Registered
Joined
·
9,884 Posts
Steve -


I don't know that Xvid is that much better than Divx, though I've heard it said the Xvid encoder might have a little better ability at motion compensation. I like Xvid mostly because it is open source (and plays faster than WM9) but I'm starting to feel that WM9, Xvid, and Divx are all at very similar levels of technology. H264/AVC is probably not quite ready for prime time but that should be interesting soon also, especially if it's approved as an HD-DVD standard.


I'm not using any of the special features of Xvid in any way, for reasons of playback performance, compatibility, and stability. I haven't found them necessary. While I've experimented some, for the most part with any new Xvid release I just click the Defaults button and leave most of the parms alone.


Except for WM9 and my one NTFS drive I don't really see advantages to XP. I haven't even tried to activate it yet. It complicates and slows my home network of Win/Me machines and I still have not tried to install my VS6 development environment. That will require VS6 service packs, mmx assembler processor pack, programmers SDK, DirectX SDK, and maybe NTDDK, the device driver SDK needed for DScaler development. And I haven't tried to install my Prassi Primo-DVD burner (from previous DVD-R drive for Win/Me) yet either.


If I find any of these is now costly or unavailable for VS6 (I've heard rumors) then I will likely just ghost my machine back to Win/Me now I've had my chance to play with WM9. The WM9 VCM codec supposed to be available for Win/Me sometime soon anyway.


- Tom
 

·
Registered
Joined
·
9,884 Posts
Quote:
MS is going to give my some stuff that trashes the XVID codec. The "green" is the masters is really the worst case senario for them. The agree that live 30FPS "green" picutre will be close if not any better. However, we will see what video they give me to try.
I didn't choose the Masters Golf clip for any "Inverse Green Lantern" scenario to make it hard for WM9. That and a Star Wars RoJ upconvert just happened to be what I was playing with at the time. IIRC, the SW clip maybe did give a very small PSNR advantage to WM9 over Xvid on 1-pass CBR. But I think it was close enough to be irrelevant.


But I'll be interested to see what sort of a clip gives the advantage to WM9 if you get one. Of course if it is just who is better at encoding film grain then I could care less. The Hollywood guys care much more about that than I ever will. ;)


- Tom
 

·
Registered
Joined
·
5,944 Posts
Discussion Starter · #20 ·
I am not sure you can compare the crapola that is WinME to the amazing engeneering effort of WinXP. I mean, in terms of stability it is not even close. The networking should be much better - ar eyou using SIMPLE file sharing or complex?


The drivers, stability, features, and speed of XP are well documneted. MY MTBF of winme is 23 hours according to MS vs 6 months to NT based systems.


To each his own - at least my games are 30% faster :)


Also, things like hyperthreading can only work on XP. ATI is not going to even support winme.
 
1 - 20 of 53 Posts
Status
Not open for further replies.
Top