Quantcast Avcodec Still Not Producing.. - digitalFAQ.com Forums [Archives]
  #1  
08-17-2004, 04:18 PM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
See these screenshots:

CCE


FreeEnc


CCE


FreeEnc


Those two frames are some 50 frames apart from each other.

The funny thing is that I would have got even better quality with CCE since I just used Q=30 and the final average bitrate would have been too low. With FreeEnc the avg bitrate was ~2500kbps and 2-pass was used. The settings were as close to each other as they possible can be and I used Inc's template in FreeEnc.

Unfortunately I have to say that the bitrate is not distributed optimally with avcodec. There's also one other scene where the difference in quality can be seen quite easily, it's a rather dark scene with ocean. Avcodec gives lots of blocks due to the dark scene and CCE gives a lot better result. I had similar results with Star Trek The Motion Picture a while back.

I wanted to post this here since there's a lot of people using FreeEnc at the moment and I didn't want to cause a stir. Maybe there's just some tweaks to do for you avcodec wizards? To me it seems that avcodec has problems with sudden high bitrate needs and in other areas (=normal daylight scenes) it really looks great.
Reply With Quote
Someday, 12:01 PM
admin's Avatar
Site Staff / Ad Manager
 
Join Date: Dec 2002
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
  #2  
08-17-2004, 05:07 PM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
Thats really strange.

Well, no contest spirit *lol* but just to show that it seems that Libavcodec isnt generally worse than CCE and that something is happen wired in your encodings ... and we should find out whats the guilty part.

I used Incredible_DVD.ini, avg2000,max7000
and at CCE min500,avg2000,max7500
all anamorph 704 x 576

As you see in my encodings its the other way around.


CCE



FreeEnc



CCE



FreeEnc



CCE



FreeEnc



CCE



FreeEnc

Reply With Quote
  #3  
08-17-2004, 07:20 PM
rds_correia rds_correia is offline
Free Member
 
Join Date: Apr 2003
Location: Chinese Democracy starts now!
Posts: 2,563
Thanks: 1
Thanked 0 Times in 0 Posts
Oh boy.
Here we go again...
I get results A and you get results Z.
I've seen this movie...
Hope this time we get it right.
I think it's time we make some testings to check the quality&speed differences between tmpg/cce and FreeENC/other libavcodec encoders.
For that we will need to agree on which options we should enable/disable in the encoders and maybe find a VOB file in the net so we all can run our tests in the same movie clip.
Anybody wants to take a step forward and point a VOB file/option pack ;
Cheers
__________________
Rui
Reply With Quote
  #4  
08-18-2004, 02:01 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Inc,

did you use 1-pass variable in CCE or a multipass VBR? My test was done with OPV.

Looking at the bitrate graph in Bitrate Viewer it looks like avcodec spends its bitrate reserve too soon, and this is exactly what happened with that Star Trek scene as well. There were lots of bits spent before the scene and then quality got a serious hit when a complex scene was encountered because there were no bits to spare in that region. That might be the reason if avcodec works like XviD : there is apparently no reserve for the whole clip; the encoder spends the bits quite close to the scene in which they were spared.
Reply With Quote
  #5  
08-18-2004, 03:32 AM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
I used 2pass in CCE as all my tests used 2pass to get a same conditions.
Ill try 1pass VBR in CCE too. But even if onepassVBR in CCE is proofed to be a bit better I never had such "big" differencies at 1pass and 2pass CCE.

Maybe you already had a look at the BV screens I did post in your "Frame loss issue" Thread. There I found out that the Q and bitrate allocation when using the Keepminrate Function was excellent.

When using 1pass in Libavcodec, exactly the problems do occur you described above - means a very bad bitrate allocation where no reserves are still present in complex scenes when encoding movies with calm beginnings and complex ends.

Gimme the exact name of that StarTrek episode, so Ill try to get that one this Weekend. Would be a nice comparison as you also do come from PAL area and maybe that movie is based on the same PAL master/origin.
Reply With Quote
  #6  
08-18-2004, 03:37 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by incredible
I used 2pass in CCE as all my tests used 2pass to get a same conditions.
Ill try 1pass VBR in CCE too. But even if onepassVBR in CCE is proofed to be a bit better I never had such "big" differencies at 1pass and 2pass CCE.
At least with OPV there shouldn't be any bitrate reserve depletions as there is no average bitrate to keep, bits are spent as they are needed. That's one possible reason for the huge quality difference in such scenes. It appears that avcodec doesn't like bright objects on a dark background. Maybe I'll try to raise the gamma in that scene and see what happens.

Quote:
Maybe you already had a look at the BV screens I did post in your "Frame loss issue" Thread. There I found out that the Q and bitrate allocation when using the Keepminrate Function was excellent
Yes, I saw them and I also used KeepMinRate() with the default values in that encode
Reply With Quote
  #7  
08-18-2004, 03:51 AM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Boulder
Unfortunately I have to say that the bitrate is not distributed optimally with avcodec. There's also one other scene where the difference in quality can be seen quite easily, it's a rather dark scene with ocean. Avcodec gives lots of blocks due to the dark scene and CCE gives a lot better result. I had similar results with Star Trek The Motion Picture a while back.
How did you produce the freenc sample ? In 2pass I guess. Because if you are comparing a Quality based output (CCE) and a 1pass encoding... that's not fair
EDIT: forget the question. Ypu already answered to it in your first post

Quote:
Originally Posted by Incredible
I used 2pass in CCE as all my tests used 2pass to get a same conditions.
I don't think this is usefull. We (almost) never use CCE in 2pass mode to do KDVD (or KVCD). And I think the only interresting thing is to compare what encoders can give to us for OUR target. It's perhaps a reality that 2pass engine of CCE is worse than Freenc. But who cares ? You are not trying to find the better 2pass encoder, do you ?
Reply With Quote
  #8  
08-18-2004, 04:04 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
And the hard part is that if you want to compare the 2-pass FreeEnc to any other encoder, you'll have to encode the whole clip, at least if you have the results I had With a quality based encode, such as CQ in TMPGEnc and OPV in CCE, you only need to encode a small clip with that problem scene in it.
Reply With Quote
  #9  
08-18-2004, 04:42 AM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
I used to do all my testing (basically when I develop new scripts) using a little making of "Star Trek : Insurrection".
It is a 5 min long clip, with mixed 4:3 and widescreen footage, slow (interviews) and fast (spatial battle) scenes.. anything I need !

YOu can find it on the PAL relase of the DVD. If you have it...
Reply With Quote
  #10  
08-18-2004, 04:43 AM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
Well as said, we (I) know that OnePas VBR does in avg result better also in CCE but not that better as like in the samples above.
Ill do that full CCE 1pass tonight and well see.

Next testobject will be LOTR 3

@ Phil
For shure I dont want to find the best 2pass encoder, Im trying to find out what results better thats all - speed is not the challange in this moment and I want to see how 1pass CCE and 2pass FreeEnc will do result in a comparison. Im not on a missionary mission but in testing phase

@generally

I dont want to force in here (as I also never did generally on libavcodec) that ITs better, but I want to know WHY some people do have -seen in a whole in avg- worse outputs - thats the thing I want to find out.
(I mean outputs in avg generally, as I know that every source behaves individual - and thats why I want to know that episode title so Ill get that StarTrek sample.)


EDIT:
@ Phil

Ill get that ""Star Trek : Insurrection"" DVD on PAL ... so its nice to know that someone gots the same testing environment (generally)
Reply With Quote
  #11  
08-18-2004, 04:49 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Speed is definitely not the issue, in fact CCE went at 1.20RT whereas FreeEnc gave me approximately 1.08RT Looks like CCE is faster on P4's, especially because the SSE2 optimizations had to be removed from FreeEnc.
Reply With Quote
  #12  
08-18-2004, 04:51 AM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by incredible
@ Phil
For shure I dont want to find the best 2pass encoder, Im trying to find out what results better thats all - speed is not the challange in this moment and I want to see how 1pass CCE and 2pass FreeEnc will do result in a comparison. Im not on a missionary mission but in testing phase
That's also why I wanted to tell you : do not do tests that are irrelevant.
As you say, you are looking for the best result and 2pass CCE is perhaps (I don't know) worse than OPV that we all use. So in other words: testing 2pass CCE mode is probably a lack of time.

Quote:
I dont want to force in here (as I also never did generally on libavcodec) that ITs better, but I want to know WHY some people do have -seen in a whole in avg- worse outputs - thats the thing I want to find out.
That is part of what I qualified as "not trustable" encoder when I spoke about mencoder. And that is why I continuously repeat that mencoder is NOT an encoder we can use (and also why I do not use it).
I don't speak about freenc that I don't even tested yet but I guess this is the same matter.
Reply With Quote
  #13  
08-18-2004, 05:44 AM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
And thats why I think it was a good decision from Boulder to post this thread in "here" - the mods area. To avoid a discussion which could again mess up parts/people in the forum.


Well CCE and Libavcodec on my Athlon do almost got the same speed, but generally I heared about a recognisable spped gain if CCE is used on SSE2 CPUs.

So this Weekend I rent from a "StarTrek addicted" friend

a) " Insurrection" and ..

b) ... the episode Boulder uses where he still owes me to tell the title
Reply With Quote
  #14  
08-18-2004, 05:51 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Hehe, the movie is Cocoon. Those frames can be found at around 38 minutes from the beginning, but my source was a TV cap so there might be a small difference. It's a scene where Brian Dennehy shows Steve Guttenberg what he's really made of.
Reply With Quote
  #15  
08-18-2004, 06:03 AM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
ok, I'll try to get that one at the video rental store.

I think you do prepare your sources using RemoveDirt(..) right?
Then there still schould be a little down to minimal amount of irritation/noise.

I just tested it this Weekend and its VERY nice and damn fast in its capability on the first view. I only had bad experiences using RemoveGrain(..) as - no matter which settings - the Details do look like osterized/blown up ... like the little TV station Logo above right.

When I do get Cocoon Im shure the mastering will be in its noise factor like a capture as its an old movie - remember my "Black rain" sample in the "how funny downscale-filter-upscale" Thread ... maybe it will do come on DVD like that.
Whe I got it, you have to send me some avisynth filtered smaples where I can exactly see in which condition lumarange/gamma/chroma/sharpness/grain/noise ... it goes into the encoders.

Totally out of Topic but I do hope the time will pass by fast as I cant wait to see what they have done to the processed StarWars Episodes DVD Collection which will be out soon
Reply With Quote
  #16  
08-18-2004, 06:29 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
I used only RemoveGrain(mode=8,modeU=2) on that one, cropping, resizing and adding borders of course. The new version includes an option to set the mode separately for chroma. The capture (in XviD!) was done by a friend of mine so I'm not sure what settings were used. They shouldn't matter anyway if you use the same source material for both encoders.

I don't know when I'll be back after the weekend since I'm getting married on Saturday, leaving home tomorrow morning, and haven't yet decided whether we'll get home on Sunday or Monday

EDIT: I'm not waiting for the SW box that eagerly, I know I'll be very pissed off to see what kind of CGI crap has been added - again. Also changing Anakin's face at the end of The Return of Jedi was something I just can't understand. They were just dancing on that poor actor's grave..
Reply With Quote
  #17  
08-18-2004, 11:43 AM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
Quote:
Originally Posted by Boulder
I don't know when I'll be back after the weekend since I'm getting married on Saturday, leaving home tomorrow morning, and haven't yet decided whether we'll get home on Sunday or Monday
Hey Boulder ....

I wish you a nice party on Saturday and all the best to you and your (future ) Wife!
Enjoy it as that seems to be much more trustable than libavcodec encodigs!
Reply With Quote
  #18  
08-18-2004, 11:52 AM
rds_correia rds_correia is offline
Free Member
 
Join Date: Apr 2003
Location: Chinese Democracy starts now!
Posts: 2,563
Thanks: 1
Thanked 0 Times in 0 Posts
Same come from here
__________________
Rui
Reply With Quote
  #19  
08-18-2004, 12:40 PM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Thanks, we'll surely have a ball. At least it's bottoms up for my friends, I'll probably try to stay out of sight and enjoy a fat cigar with a glass of whiskey. Hehe..
Reply With Quote
  #20  
08-18-2004, 10:30 PM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
Felicidades Boulder

-kwag
Reply With Quote
Reply




Similar Threads
Thread Thread Starter Forum Replies Last Post
Producing a correct 5.1 ac3 stream using free tools? (part2) J-Wo Audio Conversion 79 06-22-2005 06:48 AM
Producing a correct 5.1 ac3 stream using free tools? (part1) Dialhot Audio Conversion 15 11-04-2004 04:19 AM
Dark scenes unnatural, surfaces producing a rippling effect ? J-Wo Video Encoding and Conversion 33 06-23-2004 03:30 PM

Thread Tools



 
All times are GMT -5. The time now is 02:01 AM  —  vBulletin © Jelsoft Enterprises Ltd