Quantcast Bad I-Frame Quality with Resolution Higher than 352x288? - digitalFAQ.com Forums [Archives]
  #1  
11-04-2003, 02:12 AM
fabrice fabrice is offline
Free Member
 
Join Date: Mar 2003
Location: Madrid-Spain
Posts: 515
Thanks: 0
Thanked 0 Times in 0 Posts
Hi,

I'm trying to convert to KVCD a child movie, and in a very static scene, the I frame has a very bad quality (the source is very clean DVD), and it only affect the I frame, as the previous P frame is quite clean, and the next one has better quality than this IFrame.

previous Pframe:


bad IFrame:


This occurs with every resolution higher than 352x288, and the only way I found is make longer the GOP... I use the MA script, but occurs the same without any filters.

Anyone can help?

Thanks.
Fabrice
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  
11-04-2003, 04:47 AM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
What do you find bad in this Iframe ? Can't see on the pictures.
Reply With Quote
  #3  
11-04-2003, 04:58 AM
Krassi Krassi is offline
Free Member
 
Join Date: Mar 2003
Location: Germany
Posts: 390
Thanks: 0
Thanked 0 Times in 0 Posts
The only bad thing i can see are some tiny blocks around the crown of the praying man. And the colours are a bit different.
Is it that what you mean, fabrice
Reply With Quote
  #4  
11-04-2003, 05:33 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
@ Fabrice

Well regulary the I Frame is called "Intra" Frame where the P and B Frames refer to this. But as you use a non Standard GOP like in KVCD where the count of P & B Frames are rised up to give you furthermore an advantage in compression. And that means the GOP is already long if maxGOP in case of PAL is set larger than 15! But the KVCD matrix does compensate this, as we know.
So it would be interesting what CQ is used here! Shure, ... higher resolutions mean less CQ value.
And I would like to see 2 pics more! The frame before and after as you told that they are P & B Frames.
Cause there does appear a lot of mosquito noise at the edges within the movie but its hard to see in such a complex picture
And therefore... your I frame is not "bad" ... it just seems to much compressed becuase this part of your movie is very very complex as seen by the mpeg encoder.
Reply With Quote
  #5  
11-04-2003, 01:37 PM
fabrice fabrice is offline
Free Member
 
Join Date: Mar 2003
Location: Madrid-Spain
Posts: 515
Thanks: 0
Thanked 0 Times in 0 Posts
Hi,

I mean that the I picture is bad, because, when I saw this movie on mi TV (and this is not a HQ TV! ), I can see 'jumps' in the movie. This 'jumps' alwas occurs in a a I frame (beginning with the previous B frames), and that's what krassi says: different coulours and mosquitos' everywhere. And that's really annoying.

As far as I know, the P frame reference the previous I frame, and the B frames reference the previous and next P/I frames. The CQ was 72, but I tried with a 90 CQ, and got the same. I tried with MCencoder, and doesn't have this defect (but has others!)

The sequence is this:
PBBIBB (you have the P and the I frames).

What I don't understand is why this I frame is worst than previous frames! Even with High CQ (90), and high bitrates (max. 2500, min. 300)... I have to say that this is a long static scene... (3 sec.)

Thanks for all,
Fabrice
Reply With Quote
  #6  
11-04-2003, 02:57 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
Most importand is what GOP sequence in a whole you used!

Standard KVCD ? including 24 frames within a GOP?
Therefore 1-5823-2-maxGOP=24:
IBBPBBPBBPBBPBBPBBPBBPBB
(GOP 1-7-2 and max Gop=unlimited would give you the same result)

... and what was the gop sequence in MC Encoder? did you use this crazy 128 max gop - full of "P" frames - closed gop sequence as mentioned in the MC Thread?
Therefore: IPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPP PPP .. and so on

(BTW thats one point why in some case of using MC Encoder the quality suffers, although it compresses more!)

So we do a test using still TmpgEnc still KVCD Matrix applied just to see what happens:

Set the following GOP in TmpgEnc just for testing
1-4-2-maxGOP=unlimited
(means standard PAL gop 15)

Watch the movie (quality now is not the point, I want to know if you still get theese mentioned "Jumps"!) and tell what happens to your I frames.

Second test:
1-11-1-maxGOP=unlimited
(also means KVCD GOP 24 but! same amount of B and P frames within the sequence

Also .. watch the movie and tell me .. what happens to your "Jumps" here too.. cause these "I" frame errors should really be GOP based
And by the way what version of TmpgEnc are u using
Reply With Quote
  #7  
11-05-2003, 12:17 AM
fabrice fabrice is offline
Free Member
 
Join Date: Mar 2003
Location: Madrid-Spain
Posts: 515
Thanks: 0
Thanked 0 Times in 0 Posts
Hi,

I used the standard KVCD GOP (1/5823/2 max 25), and used the same in MC Encoder (and not the 128P GOP).
I upload a 3,2 Mb sample http://coutadeurf.en.telepolis.com/sample.m1v, ao that you can see this 'pulsations': done with CQ 75, min 300, max 2500, and the following script:
Code:
MPEG2Source("D:\dvd\mipeli.d2v")
# 
#
#BlindPP(cpu=4) 
Convolution3D(1, 6, 12, 6, 8, 2.8, 0) 

MergeChroma(blur(0.85)) 
MergeLuma(blur(0.1)) 
undot()
GripCrop(528, 576, overscan = 2) 
GripSize(resizer="LanczosResize") 

SwitchThreshold = (Width<=352) ? 4 : (Width<=480) ? 3 : 2 
ScriptClip("nf = YDifferenceToNext()"+chr(13)+ "nf >= SwitchThreshold ? \ 
UnFilter(-(fmin(round(nf)*2, 100)), -(fmin(round(nf)*2, 100)) ) : \ 
TemporalSoften( fmin( round(2/nf), 6), round(1/nf) , round(3/nf) , 1, 1) ")
undot()

#gripborders()

function fmin(int f1, int f2) { 
  return (f1<f2) ? f1 : f2 
}
(but as I said before, occurs with the d2v file, without any script)

@incredible: I did the 2 test you said, and it produce the same 'pulsations' (the first one more frequently, because of the shortest GOP). I have to see it on the TV, to see if it's visible with the 15 GOP, but I think that it is... And in this case, the second GOP produce smaller file, than KVCD GOP, with less B frames?!

The strange thing is that if the overall encoding last 50 sec., this static part (25% of the sample) last 25 sec.!

So, right now, I'm trying to kill details, to see if it helps...

Salu2
Fabrice
Reply With Quote
  #8  
11-05-2003, 04:58 AM
muhali3 muhali3 is offline
Free Member
 
Join Date: Sep 2003
Posts: 134
Thanks: 0
Thanked 0 Times in 0 Posts
Do the "pulsations" look like blocks gaining brightness and then fading away, and then doing it all over again. I get that sometimes at scene changes, but not in the middle of the movie. I took off scene change detect to fix it and my movie is looking good. hope that helps a little.
Reply With Quote
  #9  
11-05-2003, 09:32 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
Fabrice ....

I'm at work so I used Quicktimeplayer to watch several times your sample ... there are no Jumps! ....but when the camera zooms out, the image seems to swim! That could be a problem of bad filtering settings in AVS or as you right assume a problem within the GOP frame/referings and compressions

So the problem does not exist on several framesteps it exits in the whole movement.

There where several things I didn't understand in your last posting.
So now, ... follow the questions and answere them number by number..
1. .. does this effect only appear AFTER encoding?? (sorry Im not shure if I did you understand right)
2. .. if this effect also appears before encoding (previewing .avs in Vdub)
do you have the same Problem when using an .avs without filters, just a:
Code:
MPEG2Source("D:\dvd\mipeli.d2v")
GripCrop(528, 576, overscan = 2) 
GripSize(resizer="BicubicResize") # !!!!
gripborders()
????
3. If you don't see any Probs by previewing this script in Vdub, does the Imagesequence look better when encoding to mpeg using THIS script and TmpgEnc??

And what was the "second" GOP u used as reported in your last posting? Was it the "1-11-1-maxGOP=unlimited" = IPBPBPBPBPBPB..etc.... ??? Did I understand this right?

4. What do you mean in here?:
Quote:
The strange thing is that if the overall encoding last 50 sec., this static part (25% of the sample) last 25 sec.!
Does it mean the movement-scene takes longer to encode than the final static scene?

A last request ... could you just give me a few second sequence of your untouched source, just opened in Vdub and safed for example in mjpg codec using high quality?! I have to see the clear source

PS: BWT, do you mean the static part at the end which suddenly gets sharper and clearer? Thats caused by the MA routine cause the zoom-out movement stops - the NF value falls down immideately and the soften-subroutine stops.
Reply With Quote
  #10  
11-05-2003, 11:19 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
Hi fabrice,

The problem is related to the complexity of that scene (scenes) the encoder is processing. The "pulsing" you see, is the new "I" frame being inserted. For this type of movie, have you tried a shorter GOP
Maybe 1-3-2-1-10

-kwag
Reply With Quote
  #11  
11-05-2003, 12:30 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
But when using a shorter GOP he will even receive less compression and therefore les quality I frames

Well the only pump I see (maybe its my DTP Screen at work?) is when changing from motion to static...?!
The other swimm effects could also be a result of bad deinterlaced source (not done by fabrice as he has no deinterlacing lines in his script, maybe of someone before on the source?!)

But I hope we'll see the same seconds of the source
Reply With Quote
  #12  
11-05-2003, 03:07 PM
fabrice fabrice is offline
Free Member
 
Join Date: Mar 2003
Location: Madrid-Spain
Posts: 515
Thanks: 0
Thanked 0 Times in 0 Posts
Hi,

Quote:
Originally Posted by incredible
There where several things I didn't understand in your last posting.
Sorry for my bad english

Quote:
Originally Posted by incredible
1. .. does this effect only appear AFTER encoding?? (sorry Im not shure if I did you understand right)
Yes. And as I said, it don't appear at 352x288.

Quote:
Originally Posted by incredible
2. .. if this effect also appears before encoding (previewing .avs in Vdub)
The avs, like the d2v, doesn't have this problem (I even tried to encode the d2v file, and I also get this problem).

Quote:
Originally Posted by incredible
3. If you don't see any Probs by previewing this script in Vdub, does the Imagesequence look better when encoding to mpeg using THIS script and TmpgEnc??
I encoded the d2v file directly, and got the same problem.

Quote:
Originally Posted by incredible
And what was the "second" GOP u used as reported in your last posting? Was it the "1-11-1-maxGOP=unlimited" = IPBPBPBPBPBPB..etc.... ??? Did I understand this right?
Yes

Quote:
Originally Posted by incredible
4. What do you mean in here?:
Quote:
The strange thing is that if the overall encoding last 50 sec., this static part (25% of the sample) last 25 sec.!
Does it mean the movement-scene takes longer to encode than the final static scene?
No. I mean that the the final static scene takes longer to encode than the movement scene.

Quote:
Originally Posted by incredible
A last request ... could you just give me a few second sequence of your untouched source, just opened in Vdub and safed for example in mjpg codec using high quality?! I have to see the clear source
Well, I get a 46 Mb file (with 75 quality level)... Is it too big?

Quote:
Originally Posted by incredible
PS: BWT, do you mean the static part at the end which suddenly gets sharper and clearer?
I see it on my TFT monitor, but not on the TV, and I see this problem on frames 240, 264, 288, ...

@kwag: I tried with this GOP, and the pulsing is less important (there is less diference between the I and the P frames). I have to see the effect of this GOP on the file size and on the CQ, and see if the overall quality is good or not.

Thanks to everybody. I'll post the result, after encoding the whole movie with those changes.

Salu2
Fabrice
Reply With Quote
  #13  
11-06-2003, 01:57 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 fabrice
Quote:
Originally Posted by incredible
4. What do you mean in here?:
Quote:
The strange thing is that if the overall encoding last 50 sec., this static part (25% of the sample) last 25 sec.!
Does it mean the movement-scene takes longer to encode than the final static scene?
No. I mean that the the final static scene takes longer to encode than the movement scene.
Thats also what I unerstand, sorry typed wrong *lol*



Well that's really strange! What about other sources, same problems in complex scenes? Maybe a problem in TmpgEncs code? hmmmm
Reply With Quote
  #14  
11-06-2003, 04:25 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
Well that's really strange! What about other sources, same problems in complex scenes? Maybe a problem in TmpgEncs code? hmmmm
That is not stange, taht is MA script !

Unfilter is used in moving scene where temporalsoften is used in static ones. And a spatial only filter is always much faster than a temporal one.
Reply With Quote
  #15  
11-06-2003, 05:28 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
Dialhot ... my last line was not very well spearated to the line before, thats why you think Im still refering to the encoding time.

But the last line of my script was again refering to the "image" problem, say the I frame errors. Thats what still seems strange to me
Reply With Quote
  #16  
11-06-2003, 06:06 AM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
Okay
Reply With Quote
Reply




Similar Threads
Thread Thread Starter Forum Replies Last Post
PAL dimensions (352x288), but an NTSC frame rate? magwell Video Encoding and Conversion 5 12-28-2003 02:16 PM
KVCD: Higher RESolution or Higher CQ? VORTECH Video Encoding and Conversion 5 12-16-2003 10:43 AM
KVCD: Higher CQ or higher resolution? Ryo Video Encoding and Conversion 2 11-16-2003 08:23 AM
KVCD: Can we use higher resolution with kvcdx3? big Video Encoding and Conversion 2 04-26-2003 11:45 AM
KVCD: Higher resolution, but still good quality! twiztid Video Encoding and Conversion 0 04-29-2002 11:02 AM

Thread Tools



 
All times are GMT -5. The time now is 01:18 AM  —  vBulletin © Jelsoft Enterprises Ltd