Size vs Quality

[SplinterMgs]

Hi all,
I consider myself a newbie when it comes to IC and D2One and I know the issue of quality comparisons between them has been brought up before many times.
Ok, here's my situation with Road to Perdition (112 mins). I first used IFOEdit to create the main movie and new IFO files (no menus and no extras). I used IC (oversize, bumped up to about 95%) and the resulting file size was about 3.xG! I then tried D2One and its resulting file size was 4.36G!
So, what I'd like to know is whether size has direct relationship with final video quality? Does bigger size means better quality? Could IC be doing better compression (good quality but smaller size)?
I've read in this forum that IC is pretty unpredictable with final file sizes (obviously varies depending on movie content) but D2One seems to produce pretty consistent file sizes.
If I'm only keeping the main movie by using IFOEdit first, how else could I judge which produces the better quality result without sitting through and viewing both versions?
Thanks in advance for your feedback.

[hakko504]

Bitrate does not equal quality, but they are very cosely correlated, i.e. you need high bitrate to get good quality, but if you use the bits wrong you can still end up with a mess. You could say that the quality is limited by the bitrate you use.

AviSynth has a built in compare (http://www.avisynth.org/index.php?page=Compare) command that could be used for deciding which app that gives you the best quality.

(If you find the following confusing, read some of Doom9's guides on how to use Gordian Knot)

First use DVD2AVI to create 3 projects, one for each version of the movie. (use Force FILM if you have NTSC video, at least you will speed up the process by 20%, even though you don't compare all frames)
Then play this script once in the media player, without pauses, stop/restarts or anything, just once from start to stop.LoadPlugin("mpeg2dec.dll")
original=mpeg2soource("original.d2v") #This is the .d2v created from the original vobs
dvd2one=mpeg2source("Dvd2one.d2v") #This is the .d2v created from the DVD2One vobs
ic7=mpeg2source("ic7.d2v") #This is the .d2v created from the DVD2One vobs
compare(original,dvd2one,"","dvd2one.log")
compare(original,ic7,"","ic7.log")
return(original)
You can then open the logfiles in Excel and get a mathematical decision on which looks best. (sum, sum-of-squares, mean,RMS, whatever you think is appropriate)

[jdobbs]

@hakko504

That is way cool... I didn't even know that function existed in AVISYNTH. Now I can once and for all look at some objective numbers!

[jdobbs]

@hakko504

Did a quick test of DVD2one compared to InstantCopy 7. I ripped one of the Austin Powers movies to disc. I then used IC7 and created the movie. Next I used DVD2one and manipulated it so the output of the movie-only portion would be the same size as the one created by IC7. That way I could be sure that the comparison was fair because it was given at a consistent mean bitrate. Then I ran AVISYNTH's compare function to determine the PSNR found when comparing the original to each of the dups. In this test I only looked at 10,000 frames. I'll do another test on an entire movie -- I just wanted to see how it worked. Here's the results of the first test:

DVD2one (version 1.0.2):

Minimum Average Maximum
Mean Absolute Deviation: 0.0000 0.6989 1.8414
Mean Deviation: -0.1358 +0.0322 +0.3419
PSNR: 37.7004 49.9155 106.5268

InstantCopy 7.0 (version 7.0.114)

Minimum Average Maximum
Mean Absolute Deviation: 0.0000 0.6419 1.3826
Mean Deviation: -0.0435 +0.0113 +0.1097
PSNR: 40.9589 47.2933 106.5268

Interesting that in this preliminary sample it appears the at a given mean bitrate (same filesize) DVD2one seems to outperform InstantCopy as I would consider "Average" as the most important metric -- that contradicts what a lot of the subjective reporting says. It does seem that there may be a little more "swing" in DVD2one, though.

[mrbass]

That makes about as much sense to me as the first 5 equations of this (http://www.cg.tuwien.ac.at/~theussl/DA/node11.html)
Not to say it doesn't mean anything just that is over my head. Can someone please explain what this stuff means in layman terms? What was the total size of the main movie excluding any audio or menus, etc.? Less than 5GB most likely. You must use extremes to show the difference ...like longer movies.

[jdobbs]

Nothing is perfect for measuring quality. But PSNR is the best available and most widely accepted standard that can be used independent of subjective "I like this one" measurements -- especially for measuring MPEG. It measures the difference between a reference signal (the baseline) and the results of the compression (or any other action on the stream, for that matter).

Here's a link that shows one method of computing it:

http://bmrc.berkeley.edu/courseware/cs294/fall97/assignment/psnr.html

As for the size, you have to put the bar somewhere... if you raise the bitrate high enough you also wouldn't see any difference between ReMPEG2 and CCE... but we all know the difference is dramatic.

In this example a 94 minute movie was shrunk into a 2.4GB file. I used that size because it was what was created by by InstantCopy 7 when using the 7%/8% method for extras/movie on this DVD. It has quite a few extras.

I have this running against the entire movie right now -- it may come out differently... but I only trust mathmatics to determine the real quality winner. Opinions, as you know, are like as**oles, everybody has one and nobody cares about anyone elses.

[jdobbs]

@mrbass

By the way. I love your site and InstantCopy guide. Nice work.

[jdobbs]

Okay, for what it's worth -- here's the results of the full movie which is a total of 135,942 frames. PSNR is Peak Signal to Noise Ratio:

DVD2one:
Minimum Average Maximum
Mean Absolute Deviation: 0.0000 0.5176 4.7372
Mean Deviation: -0.2253 +0.0258 +0.5339
PSNR: 30.3653 50.0014 106.5268

InstantCopy 7:
Minimum Average Maximum
Mean Absolute Deviation: 0.0000 0.5886 1.3895
Mean Deviation: -0.1095 +0.0102 +0.1173
PSNR: 40.9297 47.4991 106.5268

If you only looked at this data you might get the impression that DVD2one has an advantage (as I did in the small sample) of an average of about 3dB. But, when I saw the 10dB difference between the minimum PSNR values I knew I had to look closer. So I graphed the entire stream (attached).

As the picture shows, the higher average PSNR was skewed upward by bursts of very high PSNR (probably in the easy parts of the movie). That generally isn't a good thing, as it seems to indicate that DVD2one isn't allocating its bandwidth as efficiently as InstantCopy. The data also seems to agree with the subjective reports that DVD2one's quality is "in and out" -- as you can see 20 and 30 dB swings.

I'll let the reader decide which is better in his/her mind. But I'd have to say that in my opinion this test confirms an edge for InstantCopy.

[MackemX]

@jdobbs

what was the original movie filesize? and was that last image posted for the 94 min movie which was compressed to 2.4GB?

if this is the difference on a 94min movie, what's the difference gonna be like with a 2 hour movie?. The difference in results would be greater again presumably?

very informative posts but I always knew that IC was better one for one file allocation but this puts the icing on the cake :)

can I be as bold to ask if I can I see a DVD95Copy and a IC comparision now please (only kidding :D)

cheers

p.s. it doesn't matter about the size, it's the way you use it that matters ;) :p

[jdobbs]

The movie was 3.8GB on the original disc and yes it was the 94 minute movie. So it was shrunk to ~63% of the original.

Actually I was planning to do a DVD95Copy version as well, but as you know it will be harder to make it match the filesize (it allocates very low amounts for extras and menus).

I will also do a CCE comparison.

[MackemX]

cheers

look forward to the results. In the new version of DVD95Copy I think you can adjust the reduction settings

but for now if you took a full DVD with quite a few extras to use some of the 4.37Gb when using DVD95Copy with it and see what main movie video only size you get from it

say you got a size of 3.2Gb for main movie video file then you could run IC to predict the movie title video to something like 3.42GB (including 7% error to give around 3.2Gb), and just untick all other Titles and audio to decrease processing time

then use DVD2One and CCE to get the same 3.2Gb main movie video only resulting filesizes also

with this method you could get all 4 to produce a video only file of around 3.2Gb (or whatever DVD95Copy produces firstly) to compare, that is unless you have started :)

[SplinterMgs]

Thanks for all the input. I did try out the AviSynth compare script, ran it through media player from start to finish but for some strange reason I didn't get any stats in the logfile (file exists but no numbers)! Any ideas what could be wrong?

Anyway, the results seem to suggest IC gives more consistent quality output (available bandwidth spread more evenly). But I still have an issue with Road to Perdition. As I used the movie only option with IFOEdit and created new IFO files, IC and D2One would both be compressing just the main movie. The results as I've pointed out earlier is that IC is producing a final size of 3.41G (slider at 99.9%) while D2One is at 4.36G!

Given roughly the same final file size in your tests, IC produced the better result. For movie only situations I normally try and adjust IC to produce close to 4.37G (if I can't then I use D2One) and go with it, but for this particular title it seems to be severely undersized maybe due to the many dark scenes.

So, the question remains for this particular title, could the undersized IC version still be better quality than the D2One version?

Thanks once again.

[jdobbs]

I found that in order for it to work you had to do one file at a time. If you comment out one of the "compare" lines in the script above is starts working.

[mrbass]

Originally posted by MisterX

can I be as bold to ask if I can I see a DVD95Copy and a IC comparision now please (only kidding :D)


p.s. it doesn't matter about the size, it's the way you use it that matters ;) :p [/B]

Yeah...I'm working on one right now...gonna do LOTR Extended edition...just one dvd though..not both. Was gonna do Matrix but had some "issues" we won't go into. I want to compare all three programs doing movie only ...screw the extras.

Please phone my wife and explain that one to her.

Did jdobbs just saying something about buttholes and then express an opinion or am I deaf? :confused:

[SplinterMgs]

Thanks jdobbs, but I forgot to mention that I did actually comment out the 2nd compare and got the logfile below, so what could be wrong?
Could someone please shed some light on the big difference between IC and D2One on Road to Perdition???

[Ok, I've just checked both outputs using PowerDVD with bitrate display - it seems that IC is producing lower bitrates generally due to the many dark scenes for this title! Could someone else please confirm this?]

LoadPlugin("d:\plugin\mpeg2dec.dll")
original=mpeg2source("c:\RoadToPerdition\MovieOnly\original.d2v") #This is the .d2v created from the original vobs
dvd2one=mpeg2source("c:\RoadToPerdition\RTP.d2v") #This is the .d2v created from the DVD2One vobs
#ic7=mpeg2source("ic7.d2v") #This is the .d2v created from the DVD2One vobs
compare(original,dvd2one,"","c:\RoadToPerdition\dvd2one.log")
#compare(original,ic7,"","ic7.log")
return(original)

Comparing channel(s) YUV

Mean Max Max
Absolute Mean Pos. Neg.
Frame Dev. Dev. Dev. Dev. PSNR (dB)
-----------------------------------------------------

Total frames processed: 0

Minimum Average Maximum
Mean Absolute Deviation: 0.0000 -1.#IND 0.0000
Mean Deviation: +0.0000 -1.#IND +0.0000
PSNR: 0.0000 -1.#IND 0.0000

[jdobbs]

Take out the last two parameters and see if it works on-screen.