Great! Just had a Beamish Irish Stout. Its 1:30pm here....

__________________
Christopher
Linux is the body, Unix is its soul

I'll settle for a "Samuel Adams" later on

-kwag

I've got some SA Cream Stout, Love it. Have your tried their Lite beer yet?

__________________
Christopher
Linux is the body, Unix is its soul

Never heard of that one

-kwag

I'm working on my fourth Meister Brau right now and it's 11 AM here- just kidding!

Hey Kwag,

have you tried Lanczos resizer with the LBR template? I'm really blown away by the sharpness of it. Lanczos is built into avisynth 2.06 and is a resizing option in the latest FitCD too

thanks,
ren

Great result, kwag! I'd have a beer right now, if there were any in the house .

This might be a little premature, but I'm getting some promising results with 50/48 and scene change detection OFF for this 1h59m movie...

I forgot to try that in my script. I'll run another prediction now to see how the sample comes out on "The Green (kilometric )Mile". I'll post result here

-kwag

Here, have a virtual beer

I always leave scene change detection OFF, because I've seen TMPEG insert false I frames where there's no actual scene change. So I just leave it off. Same with "Soften Blocks". It works in reverse as advertized. If you turn it on and set 100-100, you see sharper macroblock edges. I've reported this to Pegasys, Inc, and they never responded

-kwag

I might be mistaken here, but I'm pretty sure it's enabled in the x3 template. Might want to check that when you release with the new GOP structure.

I just finished sampling with a new technique I'm using (basically the same kinda deal, just slightly more accurate I think) and will know in 3h34m how close it is...

By the way, do you know any details on exactly how much overhead there is on a VCD? VCDEasy seems to use a figure of muxed_size + around 11mb for its prediction.

Turn it OFF . And sorry about that. ALL templates should be set to OFF. I probably screwed up when I made the last update.
We'll be here drinking beer waiting for your result
That number is about right. I don't have any references, but ~10 to ~11 MB is the fixed overhead structure on a VCD. That total will always be added to your mpeg files, no matter what size they are.

-kwag

Bah :P. I'm going to call my wife and tell... er, ask her to pick up some on the way home .

That explains why I had to overburn my last "100% accurate" encode .

If my current test proves to be on the money I think I might have come up with an even more accurate way of doing it that shouldn't require an error margin (or, if it does, it should be very small). Basically instead of doing calculations in seconds I'm doing it with frames:

Target media = 80 minutes = 836,640,000 bytes
Frames in movie = 171,774
Audio size = 114,633,253 bytes
Padded audio size = 115,628,333 bytes
Sample points = 50
Sample length = 48 frames
Frames in sample = 2424 ( still don't know how SelectEvery arrives at this figure... it also gives 4848 for 100/48... any ideas? If we can't come up with a formula to duplicate its frame count I'll have to look at the source, or ask dividee )

Compensate for overhead:
836,640,000 - 11,534,336 = 825,105,664
Compensate for system stream:
825,105,664 * (2018 / 204 = 813,019,155
Compensate for audio:
813,019,155 - 115,628,333 = 697,390,822

Final encode size = 697,390,822 bytes
Bytes per frame = 4,060
Target sample size = 4,060 * 2424 = 9,841,440 bytes

Theoretically this should bring us as close as possible to the correct result. I have a feeling that disabling scene change detection will help a great deal, and might explain why you always get more accurate predictions than me .

Does anyone see a problem with my theory/math?

Hi SansGrip and Kwag,

@SansGrip

I believe this is true for VCDEasy, but Nero's padding is larger.

@Kwag

I have been using the script below and have a question about using
multiples of 30 for each hour of movie time (i.e. 2hrs * 30 = 60)?
Minority Report is 139 minutes or 2 hours 19 minutes.
I see some scripts using 50????? What's the rule and are you still
trying to determine the factor being used (i.e. 0.95, 0.98, 0.95555, etc.)

LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\MPEG2DEC\MPEG2DEC.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\BlockBuster\BlockBuster.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\LegalClip\LegalClip.dll")
mpeg2source("D:\Temp\movie.d2v")
#
IL = Framecount / 60
SL = round(Framerate) * 2
SelectRangeEvery(IL, SL)
#
LegalClip()
LanczosResize(496,44
Blockbuster( method="noise", detail_min=1, detail_max=10, variance=1, seed=1 )
AddBorders(16,16,16,16)
LegalClip()


-black prince

Hey SansGrip,

Just did a test file prediction using Tmpgenc's Clip Frame. The test file
size using Clip Frame is 18,483,997 and without 20,305,188. I believe
you felt there should be no significant decrease in file size just by
masking borders. This shows a 2MB decrease with no change in
picture quality. What could be happening during Tmpgenc's mask
that accounts for this decrease

-black prince

I'm really not sure. If your Avisynth script is correct, theoretically there should be no difference whatsoever. Could you post your complete script?

Edit: I'm glad you brought this up again. I really want to thrash this out, because that's a serious gain in compressibility...

Hey SansGrip,

LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\MPEG2DEC\MPEG2DEC.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\BlockBuster\BlockBuster.dll")
LoadPlugin("E:\DVD Backup\2 - DVD2SVCD\LegalClip\LegalClip.dll")
mpeg2source("D:\Temp\movie.d2v")
#
IL = Framecount / 60
SL = round(Framerate) * 2
SelectRangeEvery(IL, SL)
#
LegalClip()
LanczosResize(496,44
Blockbuster( method="noise", detail_min=1, detail_max=10, variance=1, seed=1 )
AddBorders(16,16,16,16)
LegalClip()

I ran this script using CQ_VBR=11 and without Clip Frame file size = 20,305,188
With Clip Frame file size = 18,483,997. This is KVCDx3 (528x480)

-black prince

I notice you're not cropping in the script. How big are the "black" borders around the original, non-resized frame?

@SansGrip,

I believe LanczosResize will automatically crop to resize When I
view this script with Tmpgenc Clip Frame the borders for masking
seem correct for the aspect ratio from FitCD (16,16,16,16) or
TV-overscan #2.

-black prince

The Avisynth resizing filters only resize -- none of them crop. Make a simple .avs with just the following:

and load it into TMPGEnc. Use the mask dialog to figure out the size of the existing borders on the frame, then let me know what they are and I'll post an updated version of your script to crop them in Avisynth. I think this might be what's causing the difference.

The results are in . Untouchables, 1h59m, new GOP, no scene-change detection, 50/48.

Target file size of video: 706,804kb
Actual size of video: 697,068kb
Difference: 9,736kb = 1.39%

Not bad -- looks like I can go ahead and code this new method up and then play a little more with the sample points/sample length settings...

How does "Untouchables" look SansGrip

-kwag