Shrink 3.2 vs. InstantCopy 8 vs. CCE. Guess who won? It wasn't Shrink...

[JFerguson]

Related thread here:

my old thread

My computer crashed after that comparison, so I never got around to posting some snapshots online. I have some today, though. I'll repeat most of that thread here:

The Test:

See how a number of transcoders/encoders stack up against each other.


The Contenders:
. The original DVD
. DVD Shrink v3.1.6 w/ Deep Analysis
. DVD Shrink v3.2 w/ Deep Analysis and no AEC
. DVD Shrink v3.2 w/ Deep Analysis and AEC Sharp
. DVD Shrink v3.2 w/ Deep Analysis and AEC Smooth
. InstantCopy v8
. Big 3 (CCE SP v2.5) 4-pass (?? - read more on this below)


The Platform:
. AOpen AK79D-1394
. Barton 2500+ overclocked to 3200+
. 512MB RAM


Preparation:

I used DVD Shrink v3.2 to create an UNCOMPRESSED movie-only backup of "The Matrix Revolutions". Disc 1 of the R1 release is basically the movie, 550MB in menus, and 230MB of trailers.

With just the uncompressed movie, english soundtrack and english subtitles, the size came to 5.28GB.


Compression:

DVD Shrink v3.1.6: Compressed movie at about 81%, full analysis was run prior to compression. Final output size: 4.37GB. Running time: about 30 minutes total.

InstantCopy v8.0.0.270 (aka 8.0.5): InstantCopy shows estimated compression of about 81%. The movie is 2.40:1, so I knew the final output size was unpredictable, but it came in pretty close. Final output size: 4.25GB. Running time: about 90 minutes, I think.

DVD Shrink v3.2: Compressed movie at about 81%: full analysis was run prior to compression. Final output size: 4.35GB. Running time: about 30 minutes with no AEC, about 45-50 minutes with AEC.

Big 3: DIF4U says 4-pass but BATCHCCEWS runs ROBA template which seems to run like 2-pass (ROBA 1st part (w/ 1 CCE pass), then ROBA 2nd part (w/ 1 CCE pass)). I think I've observed different, more correct behavior than this when using CCE 2.6 and obove. For this test, it was CCE SP v2.50, though. Final output size: 4.36GB. Running time: about 5-6 hours.

Note: These compressions bring the video down from the original bitrate of around 5000 down to 4000.


Analysis:

I loaded up the original into PowerDVD v4 and captured snapshots at Chapter stops 1-32 + 2 frames. The last time I ran this comparison, I couldn't just use exact chapter stops, because Shrink v3.1.6 snapshots were bytewise identical to the original source snapshots. When, I went past the chapter stops by 2 frames, this bytewise-equivalent behavior from Shrink ceased. I don't know if Shrink v3.2 still does this (I doubt it though, when using AEC).

Using this method, all snapshots were bytewise inequivalent to the original.

The snapshots were renamed from the default PDVD_000.BMP, PDVD_001.BMP, ...:
. PDVD_000orig.BMP - The original DVD
. PDVD_000s31.BMP - DVD Shrink v3.1.6 w/ Deep Analysis
. PDVD_000s32a0.BMP - DVD Shrink v3.2 w/ Deep Analysis and no AEC
. PDVD_000s32a1.BMP - DVD Shrink v3.2 w/ Deep Analysis and AEC Sharp
. PDVD_000s32a2.BMP - DVD Shrink v3.2 w/ Deep Analysis and AEC Smooth
. PDVD_000ic.BMP - InstantCopy v8
. PDVD_000cce.BMP - Big 3 (CCE SP v2.5)

Fired up ACDSee Classic v2.43. In general, with the files being renamed as above, they were positioned so that I could PAGE UP and PAGE DOWN in ACDSee's slide show viewer in this order: first a transcoder/encoder snapshot, then the original snapshot, and finally a second, different transcoder/encoder snapshot. I may have temporarily changed the naming conventions above to keep the original DVD snapshots in the middle (like CCE vs. InstantCopy, I renamed the IC snapshots from *ic.BMP to *pic.BMP). This made it easy to switch back and forth from the original snapshot, and either of the two compressed snapshots.

When using the ACDSee slide show, I did one Zoom +, to magnify the snapshots once, and a Zoom Lock, to keep them magnified. When paging back and forth between the snapshots, I looked for what I'll call "pixel bounce". That is, when QUICKLY paging back and forth between an original DVD snapshot and a compressed snapshot, you can see the difference between the two by observing the change (or motion) in individual or clusters of pixels between the two snapshots.


Conclusion:

With this test and these samples, InstantCopy pretty consistently maintained the quality of the original frame. DVD Shrink v3.1.6 was less consistent, and in some high-motion frames, really hammered things. DVD Shrink v3.2 with AEC improved on this tremendously, but still lost out to InstantCopy on the more difficult frames. As for my 6-hour CCE encode, I think it was edged out by InstantCopy; the CCE encode really added a grainy (non) quality to some frames.

I'm posting some links here to the snapshots and tools:
. Minimum packet of snapshots (6MB)
. Additional packet of snapshots (17MB)
. Tools (3MB)

Note: These are self-extracting archives.

The MINimum packet contains snapshot #s 15, 19, 27. Add 2 to these to get the chapter stop (I started at chapter stop #3). These produced the greatest differences in quality among the tested transcoders/encoders. That made these the hardest frames to compress.

The MORe packet contains snapshot #s 1, 3, 5, 11, 14, 17, 24, 26, 28. Same thing as above to get the chapter stop. These produced various degrees of quality among the tested transcoder/encoders.

The Tools packet contains:
. ACDSee Classic v2.4.3. I thought this was free at one time, but it's shareware/nagware. You can extract it anywhere and run it.
. Beyond Compare v2.2.5. Install for a very nice Norton Commander clone (shareware). If you need to rename snapshots, load this guy up and point the two sides to your snapshot directory. Enter a filter in the filter box (like *ic*, for InstantCopy). Highlight the filtered files, and hit F2 to bulk rename them to whatever you want. This is a very nice file/directory comparison tool.


Conclusion (More Detail):

DVD Shrink v3.2 (Deep Analysis, no AEC) beats DVD Shrink v3.1.6 (Deep Analysis). Look at snapshots 15, 19, 27.

InstantCopy v8 beats DVD Shrink v3.2 (Deep Analysis, AEC Sharp or AEC Smooth). Look at snapshots 15, 19, 27. In addition: 14, 17, 24, 26, 28.

DVD Shrink v3.2 (Deep Analysis, AEC Sharp) vs. DVD Shrink v3.2 (Deep Analysis, AEC Smooth). A toss-up maybe. Look at 15, 19, 27. In addition: 1, 5, 11, 26. I dunno, maybe Sharp wins out here -- look at 19.

InstantCopy v8 beats CCE SP v2.50 -- HUH???! Look at 19, 27. In addition: 3. Look at others too, things seem grainy with the CCE encode.


Closing remarks:
. I was surprised by the CCE vs. IC results.
. DVD Shrink v3.2 did a pretty good job given it's low execution time.
. I wonder how things might work out on a title needing higher compression (like 40% versus this test's 20%).

Hmmm...

[luphy]

I believe DVDShrink's AEC algorithm only kicks in at certain compression levels.

So your test sample may not have had it kick in most of the time....no idea.

I know that the author said that choosing Max Smoothness will essentially turn on the AEC algorithm full-time.

So would be interesting to see the result of a similar comparison at high compression as you suggested. Or maybe redo one AEC run using Max Smoothness.

Thanks for your thoughts. Oh, and as for IC8 beating out CCE, it's possible, in theory, for trancoders to have better results versus encoders, for low-compression projects.

[Hemmo]

Originally posted by JFerguson
The snapshots were renamed from the default PDVD_000.BMP, PDVD_001.BMP, ...:
. PDVD_000s31.BMP - DVD Shrink v3.1.6 w/ Deep Analysis
. PDVD_000s32.BMP - DVD Shrink v3.2 w/ Deep Analysis and no AEC
. PDVD_000s32a0.BMP - DVD Shrink v3.2 w/ Deep Analysis and AEC Sharp
. PDVD_000s32a1.BMP - DVD Shrink v3.2 w/ Deep Analysis and AEC Smooth

Thanx for the test, but.
I get from exe:
PDVD_003s31.BMP <- DVD Shrink v3.1.6 w/ Deep Analysis ?
PDVD_003s32a0.BMP <- DVD Shrink v3.2 w/ Deep Analysis and no AEC ?
PDVD_003s32a1.BMP <- DVD Shrink v3.2 w/ Deep Analysis and AEC Sharp ?
PDVD_003s32a2.BMP <- DVD Shrink v3.2 w/ Deep Analysis and AEC Smooth ?
I have no:
PDVD_000s32.BMP
Is that right order ?

[quantum]

They don't seem to be labelled right. I'm guessing the 32a0's are the Shrink 3.2 with no AEC and 32a1 and 32a2 are the two with AEC enabled. I also wonder why default was left out. Why include 3 different non-default versions but not the default? Personally I'd rather see only the default since that would be what most users would select.

You have to be careful with CCE as it has built in filters that affect the picture. I don't recall if version 2.5 has these active by default, but they're active by default in 2.6x. They're supposed to improve the picture and will introduce differences. You can turn them off.

You also have to be careful if you're comparing at or near scene changes. Frames around these areas may not be representative of the average.

I've found it best to interactively compare many adjacent frames as described here.

With that said, personally I think the comparisons show Shrink 3.2 and IC very close, with some frames going either way. If you have to zoom in to %200 on your computer monitor to see differences then I think it demonstrates they are substantially the same, and any differences between the two are very likely to be invisible on a TV screen.

Where IC8 and Shrink 3.2 are in all common sense "equal", Shrink wins hand down being FREE, and fast. And CCE needs an expert at the handles if it is to be compared to anything, but even there the amateur with shrink wins the amateur with CCE at compressions lower than 60% (using shrink-scale where lower means MORE COMPRESSION).

But usually to me (and as proven by many people when transcoding was new), a good 80%+ encoding with a transcoder looks better than the same with CCE. It looks more original, and lacks the additional mosquito noise added by a full encoding.

[JFerguson]

Originally posted by luphy
I believe DVDShrink's AEC algorithm only kicks in at certain compression levels.

So your test sample may not have had it kick in most of the time....no idea.

I know that the author said that choosing Max Smoothness will essentially turn on the AEC algorithm full-time.

So would be interesting to see the result of a similar comparison at high compression as you suggested. Or maybe redo one AEC run using Max Smoothness.

Thanks for your thoughts. Oh, and as for IC8 beating out CCE, it's possible, in theory, for trancoders to have better results versus encoders, for low-compression projects.

Well, there are differences in snapshots between the three AEC compressions (none, Sharp (default), and Smooth), so I know it kicked in.

It's either on or off, isn't it?

[JFerguson]

Hemmo...

I fixed the naming conventions for those snapshots. You got it right. Thanks...

[JFerguson]

Originally posted by quantum
They don't seem to be labelled right. I'm guessing the 32a0's are the Shrink 3.2 with no AEC and 32a1 and 32a2 are the two with AEC enabled. I also wonder why default was left out. Why include 3 different non-default versions but not the default? Personally I'd rather see only the default since that would be what most users would select.

AEC Sharp is the default and is included. Is there a different default?

Originally posted by quantum
You have to be careful with CCE as it has built in filters that affect the picture. I don't recall if version 2.5 has these active by default, but they're active by default in 2.6x. They're supposed to improve the picture and will introduce differences. You can turn them off.

You also have to be careful if you're comparing at or near scene changes. Frames around these areas may not be representative of the average.

I've found it best to interactively compare many adjacent frames as described here.

With that said, personally I think the comparisons show Shrink 3.2 and IC very close, with some frames going either way. If you have to zoom in to %200 on your computer monitor to see differences then I think it demonstrates they are substantially the same, and any differences between the two are very likely to be invisible on a TV screen.

Yeah, they're close, but there's no contest on the hardest frames (high movement, high action). It's InstantCopy. I don't have to Zoom in 200% to see Shrink 3.2 wash out certain elements of frames (see snapshots 15 and 19, especially AEC Smooth), it's just easier. Plus, I watch all my movies on TV's that are 50+ inches, so I don't know how invisible these would be...

[JFerguson]

Originally posted by Joergen
Where IC8 and Shrink 3.2 are in all common sense "equal", Shrink wins hand down being FREE, and fast. And CCE needs an expert at the handles if it is to be compared to anything, but even there the amateur with shrink wins the amateur with CCE at compressions lower than 60% (using shrink-scale where lower means MORE COMPRESSION).

Agreed, Shrink 3.2 has a lot going for it. Shrink 3.1 was down and out on this one, but the new version is definitely a contender. I use Shrink a lot for its simplicity and speed, but stayed away from it for anything over 15% compression (which didn't happen to often, with stripping). With Shrink 3.2 and AEC, it looks like I can relax that rule a bit.

I should try it on one of my TV DVDs (like Alias), which I usually run through InstantCopy at about 40% compression (60% original). Might make for a good test. Heck, I think I'll start it now...

Originally posted by Joergen
But usually to me (and as proven by many people when transcoding was new), a good 80%+ encoding with a transcoder looks better than the same with CCE. It looks more original, and lacks the additional mosquito noise added by a full encoding.

I didn't know that. Definite mosquito noise on this one!

[dragongodz]

as i said back in that other thread though 1 test/movie is not enough of a test to really prove anything. ok it proves it for 1 movie at 1 compression level but thats all.

waiting for the 3+ movie (all at different compression levels) test.

[begu]

Yes, I also agree that the screens has to be zoomed to 200% or more to see the differences. This is because some of us has big TV screens or projector like I have. So, it is definitely true, that minor macroblocking and other compression artifacts as shown in the computer monitor can be very irritating and not anymore that minor on 80 inch projector screen.

And I have to say that the new shrink really does nice job! I tried the LOTR II extended edition disc 2 (R2) with 63% (no DTS, no 2ch dolby audio, all other audio tracks kept and menus compressed to ~37%). I used the default for AEC. And it looked quite good, even compared to the dvd rebuilder with CCE with same 63% of the original size. There were less compression artefacts in the shrink version, but the judgement that wich is better is ahrd to say. Somehow the CCE looked maybe more natural (with noise) but again in some scenes the compression artefacts were quite irritating. So it is really personal tastes, that will decide wich one method to use. For now, I have to do the both and then decide wich one to keep.

[daehkcid]

When should you use smooth and when sharp? (and when max?)

[int 21h]

I think you will find different results at drastically higher compression levels (for obvious reasons)

[SeeMoreDigital]

Hi all,

Was dvd2one (v1.5.1) released 20 Aug 04, not even worth considering?


Cheers

[Oldeman]

I have Instant Copy, but I don't use it very much. On my P4 system DVDShrink runs in about one fourth the time, gives accurate sizes, AND is easy to use.

In my own testing, these was very little difference between Instant copy and Shrink 3.2 output.

Net: DVDShrink is GOOD, FAST, EASY and FREE....

[daehkcid]

Originally posted by int 21h
I think you will find different results at drastically higher compression levels (for obvious reasons)

What do you consider higher compression?

[Beak]

Originally posted by Oldeman
I have Instant Copy, but I don't use it very much. On my P4 system DVDShrink runs in about one fourth the time, gives accurate sizes, AND is easy to use.

In my own testing, these was very little difference between Instant copy and Shrink 3.2 output.

Net: DVDShrink is GOOD, FAST, EASY and FREE....

I agree totally with this. Its great that people are doing things like zooming and analyzing the outputs of the various programs. Always important to see for sure what is best.

But most movies I have backed up require less than 20% compression. I can't tell the diff on a 36 inch Sony XBR.

My b/u of Saving Private Ryan where I kept the DTS soundtrack looks very good.And that movie has tons of action and steady-cam sequences. This was around 45% compressed.This was with an older version of Shrink.

So I would just like to echo Oldeman's thoughts and thank those concerned for such a great free tool.

[Trahald]

Maybe its my choice of movie, but most of the movies i get need 30-~45 % compression. quantizer transcoders do ok at 20->25% but at higher % (at least to my eyes) you need full compression (cce/quenc etc).

[daehkcid]

It all comes down to time and effort..

[SciFer]

What about SeeMoreDigital question:

Hi all,

Was dvd2one (v1.5.1) released 20 Aug 04, not even worth considering?


Cheers


I would like to see everyone's feedback