Quantcast CQ vs. CQ_VBR ... Very Interesting... - Page 29 - digitalFAQ.com Forums [Archives]
  #561  
01-14-2003, 02:36 AM
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
Quote:
Originally Posted by SansGrip
I just finished encoding Death To Smoochy (new 352x480 template) at CQ 58.5. Target size was ~689mb, final size was ~648mb. I was able to bump the audio up from 112 to 160 .
Just finished "Count of Monte Cristo". Predicted size = 692MB. Actual final size = 670.8MB.
Maybe we need the factor again Maybe ~0.98 seems a good center point My encode was ~3% low, which is good. I rather have a safety margin than go over the CD-R limit. So if you got ~6% and I got ~3% without a factor, then with a factor we should be in a range of 0 to -3% accuracy, and that's not bad at all

-kwag
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
  #562  
01-14-2003, 07:12 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 SansGrip
I'm not sure, but I notice that a lot of very dark movies are hard to compress (e.g. Panic Room, Resident Evil).
Panic Room is hard to compress because the DVD is very noisy IMO Maybe the greenish picture also contributed to it being a tough case.
Reply With Quote
  #563  
01-14-2003, 07:52 AM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by kwag
Maybe we need the factor again Maybe ~0.98 seems a good center point
I'd be sorry to reintroduce that, because it's really just a hack. I'd like to know why we didn't need it with the previous settings but seem to need it now...

I'm going to do Signs today. If that's out significantly too then it might be time to reinvestigate the formula.
Reply With Quote
  #564  
01-14-2003, 09:31 AM
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
Quote:
Originally Posted by SansGrip
Quote:
Originally Posted by kwag
Maybe we need the factor again Maybe ~0.98 seems a good center point
I'd be sorry to reintroduce that, because it's really just a hack. I'd like to know why we didn't need it with the previous settings but seem to need it now...
Yes I agree with you, but what about if the factor is always constant, and we always get the same results. Maybe the factor is needed, as it may be the actual average compressibility% of the GOP for our formula This way, if there are future changes (Hope not ) of GOP size, etc, the formula can remain constant, and just adjust this offset. That is, again, if the results are always constant within a %margin. Right now, the way I see it, with a 0.98% factor we are within -2% to -3% of target. That is with the numbers you got, and my last encode on "Count of Monte Cristo"

-kwag
Reply With Quote
  #565  
01-14-2003, 10:13 AM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by kwag
Right now, the way I see it, with a 0.98% factor we are within -2% to -3% of target. That is with the numbers you got, and my last encode on "Count of Monte Cristo"
Signs will be done in 5 hours, so I'll have a better idea then.

By the way, get this: with Signs I was able to use a CQ of 73 at 528x480 as opposed to Death To Smoochy's CQ of 62 at 352x480. This despite the fact that they are almost exactly the same length and have the same aspect ratio (16:9).

Very odd. I think I shall try Death To Smoochy again with stronger smoothing. I notice some noise in there that Flux can't get rid of, so I'll try adding TemporalSoften or perhaps C3D and see what happens.
Reply With Quote
  #566  
01-14-2003, 10:22 AM
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
Incredible as it may sound, "The Green Mile" is currently being encoded at 352x240 LBR with a CQ of 62 . I'm encoding that on a Celeron 1.2Ghz (Slowwww!) and it's currently at 74% done ( Elapsed: 08:55:00, Remaining: 02:59:58 ). File size so far is 443MB. Predicted target size is 595.86MB. We'll see what the final size is

-kwag
Reply With Quote
  #567  
01-14-2003, 10:26 AM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Here's a sample of Signs encoded for one disc with 112kbps audio. I think this is about the best quality I've made yet...
Reply With Quote
  #568  
01-14-2003, 10:35 AM
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
Quote:
Originally Posted by SansGrip
Here's a sample of Signs encoded for one disc with 112kbps audio. I think this is about the best quality I've made yet...
And that's the BEST sample I've seen
What filters did you use on that

-kwag
Reply With Quote
  #569  
01-14-2003, 10:55 AM
Yoda Yoda is offline
Free Member
 
Join Date: Apr 2002
Posts: 172
Thanks: 0
Thanked 0 Times in 0 Posts
Sopranos 3rd season episodes 3-5. Constant cq 65, 169 min, 158 meg audio. I used the new mod and 352x240 for 1 disk. I am very impressed with the quality I used to have problems with objects in the background shimmering but no longer!

Heres the script:
LoadPlugin("D:\DVD2SVCD\MPEG2Dec\MPEG2DEC.dll")
LoadPlugin("D:\DVD2SVCD\MPEG2Dec\fluxsmooth.dll")
LoadPlugin("D:\DVD2SVCD\MPEG2Dec\blockbuster.dll")
LoadPlugin("D:\DVD2SVCD\MPEG2Dec\legalclip.dll")
LoadPlugin("D:\DVD2SVCD\MPEG2Dec\sampler.dll")

Mpeg2Source("D:\Movies\Vob\Sopranos\Season3_5to7.d 2v")
LegalClip()
LanczosResize(336,192,45,0,630,480)
#bilinearResize(336,192,45,0,630,480)
FluxSmooth()
Blockbuster(method="noise", variance=.5, seed=1)
AddBorders(8,24,8,24)
LegalClip()

I used lanczos because it gave me a sharper picture.

Edit: Encode time dropped to 3hr 44min. Incredible!

-Yoda
__________________
May the code force be with you
Reply With Quote
  #570  
01-14-2003, 11:20 AM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by kwag
And that's the BEST sample I've seen
Here's the full sample strip (11.7mb), if you want to see it all.

Quote:
What filters did you use on that
New x3 template, 400 min bitrate.

Code:
Mpeg2Source("..\signs.d2v")
Telecide()
Decimate()
GripFit(528, 480, overscan=1)
FluxSmooth()
TemporalSoften(2, 5, 7)
Cnr2()
#Sampler()
LegalClip()
That's a lot of smoothing, but it seems that when you encode to low-bitrate MPEG-1 the removed details are "simulated" by the slight high-frequency artifacts... Gibbs might be useful after all .

It looks nice on the monitor, but the real test will be how it looks on the TV. I'm hoping it's not too smooth, but if so I might try using a 0.33/0.33 (or even 0/0.5) bicubic instead of bilinear, or possibly an edge-enhancement after smoothing.

What I'm getting from this is that there's quite a lot more to investigate wrt to preprocessing. Maybe even some more filters to write .
Reply With Quote
  #571  
01-14-2003, 11:24 AM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Yoda
I used lanczos because it gave me a sharper picture.
Careful with Lanczos, though -- it decreases compression fairly significantly. When reducing as opposed to enlarging I sometimes find it might be a little too sharp, so give it a try with a neutral or precise bicubic and compare (neutral: a=0.33 b=0.33, precise: a=0 b=0.5).
Reply With Quote
  #572  
01-14-2003, 11:31 AM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
@kwag

Could you try that full sample strip on your HDTV? I'd love to know how it looks on a nice clear screen. I'm gonna play it on my analog when I get back from taking my daughter to school...
Reply With Quote
  #573  
01-14-2003, 11:38 AM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
A further thought -- I think the new GOP might remove the need for Blockbuster (damn you ):

While the DCT blocks are there, and clearly visible when you look for them, they're no longer dancing around. As far as I'm concerned this makes them much much less of a problem for TV viewing, since they're no longer distracting me from what I'm supposed to be looking at.
Reply With Quote
  #574  
01-14-2003, 11:39 AM
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
Quote:
Originally Posted by SansGrip
Here's the full sample strip (11.7mb), if you want to see it all.
Quote:


It looks nice on the monitor, but the real test will be how it looks on the TV.
I'm pretty sure it's going to look almost like the original
Quote:

What I'm getting from this is that there's quite a lot more to investigate wrt to preprocessing. Maybe even some more filters to write .
Let the party continue
Reply With Quote
  #575  
01-14-2003, 01:16 PM
pacodoni pacodoni is offline
Free Member
 
Join Date: May 2002
Posts: 78
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to pacodoni
Hi all

First of all congratulations for each one of you guys, the new temp and filters really made the impossible

I did a lot of encodings with all the resolutions, and the results were amazing.


Could fit easily episode 2 ( 145 min ) in the LBR with CQ 77 and 192 sound, and also in 352x480 sound 128, next step try it in 528X480. :twirl:

For who want to try MPEG-2, i've ran some tests and could notice that there are more visible artifacts, since you have to drop CQ.

I gave it a try cos i have a brother with a Pioneer DV-525, that doesn't handle 352X480 or 528x480.

Also i'll see if Gripfit can do something about this pioneer issue, so, will encode and cross my fingers.

See ya later

Pacodoni
Reply With Quote
  #576  
01-14-2003, 01:21 PM
Yoda Yoda is offline
Free Member
 
Join Date: Apr 2002
Posts: 172
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Yoda wrote:
I used lanczos because it gave me a sharper picture.


Careful with Lanczos, though -- it decreases compression fairly significantly. When reducing as opposed to enlarging I sometimes find it might be a little too sharp, so give it a try with a neutral or precise bicubic and compare (neutral: a=0.33 b=0.33, precise: a=0 b=0.5).
_________________
Regards,
SansGrip
I'm just surprised to get that length (168 min) on a single cd at such as good resolution. I don't think the extra bytes will be that much to worry about. I'm doing the encode as we speak and at 80% I'm at 512mb. Right on target to 640mb. This is also a full screen movie so I expect to be able to pack more using a 16:9 wide screen. I'm just very,very happy with the latest template and methods. You guys are geniuses!

-Yoda
__________________
May the code force be with you
Reply With Quote
  #577  
01-14-2003, 01:23 PM
Daagar Daagar is offline
Free Member
 
Join Date: May 2002
Posts: 158
Thanks: 0
Thanked 0 Times in 0 Posts
SansGrip,
For your Signs sample, your .avs makes use of telecide()/decimate(). Is there a reason why you did this instead of simply using Force FILM in DVD2AVI?
Reply With Quote
  #578  
01-14-2003, 01:52 PM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by pacodoni
next step try it in 528X480.
I think you'll find that the audio sounds just fine at 112kbps if you resample to 44.1khz (there are noticible artifacts at that bitrate if you leave it at 48khz). That gives you a bit more room for video. I recommend using HeadAC3he for audio encoding.

Quote:
Also i'll see if Gripfit can do something about this pioneer issue, so, will encode and cross my fingers.
I don't think it will. GripFit basically just replaces the Crop/XxxResize step in the script, so you'll end up with exactly the same encode as if you did it manually with FitCD. It is easier though .
Reply With Quote
  #579  
01-14-2003, 01:54 PM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Yoda
I'm just surprised to get that length (168 min) on a single cd at such as good resolution.
Especially considering it's fullscreen...

Quote:
I don't think the extra bytes will be that much to worry about.
You might be surprised how much you can increase the CQ if you switch to a bicubic or bilinear. It might be interesting to encode a sample strip of the correct size and compare them .
Reply With Quote
  #580  
01-14-2003, 01:58 PM
SansGrip SansGrip is offline
Free Member
 
Join Date: Nov 2002
Location: Ontario, Canada
Posts: 1,135
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Daagar
For your Signs sample, your .avs makes use of telecide()/decimate(). Is there a reason why you did this instead of simply using Force FILM in DVD2AVI?
Force FILM mode is great if the movie is 100% telecined, but I often find some interlaced material creeps in there too (especially during special effects or when there are captions on the screen). This can produce noticible interlacing artifacts in the final encode. Telecide will catch those strays and deinterlace them.

The disadvantage is that Telecide is quite slow: my encode times drop from on average ~3.5 hours to ~5 hours. I do think it's worth it though.
Reply With Quote
Reply




Similar Threads
Thread Thread Starter Forum Replies Last Post
Avisynth: Interesting results with YlevelsS supermule Avisynth Scripting 2 08-06-2006 11:59 PM
Avisynth: Interesting ASharp phenomenon... audioslave Avisynth Scripting 12 10-23-2003 06:36 AM
Interesting info about the Luminance Level in CCE digitalize Video Encoding and Conversion 0 04-28-2003 12:29 PM
A couple of interesting links.. kwag Off-topic Lounge 0 12-31-2002 03:47 PM
KVCD: Interesting poll found kwag Video Encoding and Conversion 2 12-31-2002 02:44 AM

Thread Tools



 
All times are GMT -5. The time now is 05:05 PM  —  vBulletin © Jelsoft Enterprises Ltd