What's Next for KVCD as a Format? - Page 5

rhino · #81 04-25-2003, 08:45 AM

It must have taken about 20 minutes to load up. Now i have a 1800XP processor. This is where an Itanium 2 would come in handy if tmpgenc was ported to 64 bit.

Then if you want to go in and adjust things, you have time to wait again. Maybe its time to consider writing our own cross platform encoder. Maybe I can then take advantage of that 8-way server sitting in the lab!!

Abond · #82 04-25-2003, 08:46 AM

May be in the data file should be:

Code:

BitRate=VBR

or

Code:

BitRate=MVBR

kwag · #83 04-25-2003, 08:53 AM

Quote:

Originally Posted by rhino

It must have taken about 20 minutes to load up. Now i have a 1800XP processor. This is where an Itanium 2 would come in handy if tmpgenc was ported to 64 bit.

Well, I guess that tells it all

Right now, I'm getting ~5 minutes to find CQ with the new ToK and the revised file prediction, so nothing beats that at this moment

Quote:

Maybe I can then take advantage of that 8-way server sitting in the lab!!

That must be a screamer

-kwag

kwag · #84 04-25-2003, 08:55 AM

Quote:

Originally Posted by Abond

May be in the data file should be:

Code:

BitRate=VBR

or

Code:

BitRate=MVBR

Well it's actually CBR, because each frame has a different fixed CBR, and the end result is VBR

-kwag

rhino · #85 04-25-2003, 08:57 AM

Quote:

That must be a screamer

Its got 4 gigs of memory as well, so i could create a ramdisk as well. But we got to write a unix version of tmpgenc first.

One thing we could do in tmpgenc is for B frames and P frames is set the relative quality to the previous frame. But if its going to take so long to load in the file and process it, then I guess its probably not worth the effort.

We would only have to load in the complete file for complete movie encode. For the samples etc. we would only need to load in a subset, we just got to work out what the subset would be,

rhino · #86 04-25-2003, 09:18 AM

If we can find out the bitrate for every frame or any other frame information, is there any way we could figure out projected file size from this information?

Abond · #87 04-25-2003, 09:52 AM

Quote:

Well it's actually CBR, because each frame has a different fixed CBR, and the end result is VBR

Yes, but we have a strange behavior of TMPG and may be it require strange data

. From the example here above is seen that it encode as CBR taking the BR from the I frame. Setting this as VBR might change this situation...

rhino · #88 04-25-2003, 10:32 AM

Unfortunately VBR field type is ignored.

Cheers,

Snowman · #89 04-25-2003, 12:27 PM

I have tested the last view days with you and I think i have a small hint ;

Maybe you change the words Framequality and bitrate, what means that you set once the bitrate and the next frames you set the quality.

I hve tested that since the last posts; it works fine- quality bad because everything is estimated .

Sorry for my bad style but i am from germany

kwag · #90 04-25-2003, 01:23 PM

Hi Snowman,

I just finished a long chat with hedix, and he gave me similar hints to what you just described. It IS possible to do something! And I think the key is exactly in the Framequality variable, and offset that value from an established bit rate. We just have to find out if, for example frame quality of 50% from a bit rate of 2,500, is exactly 1,250. Or if it's not, but it's linear, then find the corresponding linearity of FrameQuality/BitRate

The day looks a little brighter to me than earlier this morning

Thanks Snowman, your style is fine!

-kwag

rhino · #91 04-25-2003, 02:34 PM

Hmm, for a test then I'll set the Bitrate for each I frame and then I'll set the P frames and B frames relative to the preceding I frame. This will be easy enough to do.

Cheers,

rhino · #92 04-25-2003, 03:07 PM

For anyone who is interested I have generated a file with the I, P and B frames for the Matrix (with fudged bitrates), normalised against the KVCD bitrate scale, re-averaged against MovieStacker average (approx, think i used the wrong values, but seeing as its all fudge at this stage it does not matter), then calculated the FrameQaulity against the previous I frame.

Now, it might all be completely wrong, but hey it might be a laugh!

The perl script and the txt file can be obtained from

http://kvcd.atlanticisle.com/3rdparty/

BTW, it takes a while to load in the out2.txt file to force your own bitrates.

Have a good weekend all!

Rhino.

kwag · #93 04-25-2003, 03:11 PM

Thanks Rhino

I'll see if I can play with it this weekend

-kwag

ovg64 · #94 04-26-2003, 02:21 PM

I dont know but maybe this will help your project:

A Bitrate Control Algorithm for the Berkeley MPEG-1 Video Encoder

A simple bitrate control algorithm was implemented in the Berkeley MPEG-1 Video Encoder. A series of experiments are presented that illustrate the effect of bitrate control.
http://bmrc.berkeley.edu/research/pu...Rate%20Control

Snowman · #95 04-27-2003, 11:08 AM

I have tested the whole weekend, but ther are some problems :

1. tempgenc wouldn´t load the big I,p,B text file from Ifoedit

2. If i choose the I pictures everything works fine.

3. the output file has max Gop like the original DVD file maybe IBBPBBPBBPBB ( 12 gop ).

So I think the file is getting too big for us..........

I am a little bit surpised about tmpgnc gop search ; in my film " about a boy" tmpgenc find with the scene detection at 0,63,400,514,516,818,1430,1634 framenumbers an I pic.

At the real DVD vob file there are normaly : 0,12,24, etc....but thre are also some differnt ; 516,518 etc.....

Normally they didn´t compare withe tmpgenc search style...... tempgenc mostley detected an I pic 2 frames after the real... so . origin there is one I pic at frame 818 and tempgenc find it at 820......

So we should ask us a view more questions ........

After the whole weekend, serching for the big shot I think we need a program like rempeg2 mixed with tmpgenc.

rhino · #96 04-28-2003, 03:46 AM

The Fudged file I generated from Kwags modified DVD2AVI loaded into TmpgEnv no problem (though it took about 40 minutes) and it did produce interesting results.

For the Matrix it came out with 550Mb for an average bitrate of 689 at 352x288 and 630Mb for 528x576. Of course the data is not accurate and reading the Berkeley docs on mpeg information the B frame is going to be fun to work out as it is based up preceding and fortcoming frames.

If we can figure out how to create the B frame and P frame information correctly (maybe from KWags modified DVD2AVI - I am still waiting to get Dev Studio) then maybe we can get all the frame information to correctly generate a KVCD bitrate mapping of the entire movie.

I wonder if we have the total number of frames, the average bitrate, a mapping for each frames bitrate/quality and the desired resolution is there a correlation between the final size of the outputted mpeg file. There could be an alternative to file prediction?

Cheers,

black prince · #97 04-28-2003, 07:51 AM

@rhino,

rhino wrote:

Quote:

The Fudged file I generated from Kwags modified DVD2AVI loaded into TmpgEnv no problem (though it took about 40 minutes) and it did produce interesting results.

I wonder if you could print a sample of this file. I'm still puzzled about
the syntax needed for Tmpgenc to respond

-bp

rhino · #98 04-28-2003, 09:00 AM

@bp

This is the input file I have generated for TmpgEnc, with the B and P frame quality relative to the previous I Frame.

Cheers,

Code:

0,I,BitRate=CBR:807,New Group=1
1,B,FrameQuality=1.1
2,B,FrameQuality=1.1
3,P,FrameQuality=1.3
4,B,FrameQuality=1.0
5,B,FrameQuality=2.1
6,P,FrameQuality=1.8
7,B,FrameQuality=1.1
8,B,FrameQuality=0.7
9,P,FrameQuality=0.5
10,B,FrameQuality=0.4
11,B,FrameQuality=1.8
12,P,FrameQuality=1.9
13,B,FrameQuality=1.4
14,B,FrameQuality=1.9
15,I,BitRate=CBR:1505,New Group=1
16,B,FrameQuality=0.8
17,B,FrameQuality=0.3
18,P,FrameQuality=0.2
19,B,FrameQuality=0.6
20,B,FrameQuality=0.9
21,P,FrameQuality=1.1
22,B,FrameQuality=0.7
23,B,FrameQuality=0.9
24,P,FrameQuality=0.2
25,B,FrameQuality=0.4
26,B,FrameQuality=0.3
27,P,FrameQuality=1.2
28,B,FrameQuality=0.8
29,B,FrameQuality=1.1

rhino · #99 04-30-2003, 04:35 AM

TmpgEnc 2.511 seems to have been improved in the loading of force frame data file. Its done to about 20-30 minutes for for a 2 hr film. And they seem to have improved overall multi-tasking ability in windows.

Cheers,

black prince · #**100** 04-30-2003, 02:05 PM

@rhino,

Curious about FrameQuality for P and B frames in relationship to it's
associated I frame. How is the CBR applied to a P or B frame???
I have seen a sample load file that converts frame type, e.g.
1,P,@,B. Not sure how this would be useful now, but later it could be

-bp