Comparison TMPGenc and CCE with DVD2SVCD

[24hourloop]

I ran a (very loooong) test comparing 'Contact' TMPGenc 2 pass VBR with CCE 3 pass VBR. I found this a rather fair comparison and I am certain that 2 pass CCE would have yielded similar results.

I compared the resulting output files (actually, the middle CD) using Birate Analyzer and while TMPGenc is doing well, it just isn't there yet.

CCE yielded an average quantitization factor (the smaller, the better) of 1.65 whereas TMPGenc only reached 3.83. Maximum quantitizations were equally distinct: 4.98 for CCE as opposed to 10.53. This does *not* mean that TMPGenc's picture quality is twice as bad as CCE's, it is merely a scientific way of looking at the compression quality and 'smaller is better'.

Obviously the resulting output files were about the same size and TMPGenc took *a lot* longer (I refrained from timing it besides the fact that I went to sleep and it still hadn't finished until afternoon or so.

Peak and average bitrates were in the same ballpark.

Comparing allocated bitrates and resulting quantitizations between the two compressors clearly showed that CCE is *a lot* better able to determine spots needing higher and lower allocations. SOmething for TMPGenc to work on.

Nonetheless, TMPGenc represents a viable (not to speak of legal, affordable and continously developed) alternative.

I believe the effort to implement the support was worthwhile.

[da franksta]

interesting stuff! thanks for sharing it. :)

[Arianos]

Well, I did not ran such an exhaustive, technical and detailed test, but I did something similar, out of necessity:
The PC is a P4@1800 with 512 Ram.
First I encoded "Wateworld" with 1.0.6 b4. The whole process with CCE 5 pass, took exactly 12 hours. The result was the usual excellent.
Next I tried "Castaway" with 1.0.7 b1 (as it came out) and CCE froze the PC at about 72% (Second time ever - after 60+ encodings). I thought of it as a good opportunity to test drive TMEGenc to see first hand what the rave is about. I set it at 2 VBR, Highest quality. I did a crash recovery, so the audio files etc, were already there. It took TMPEGenc alone about 23 hours!!!! The quality , IMHO, leaves a lot to be desired.
Both were similar movies (length, theme) both difficult subjects (water scenes) both with high bitrate settings (max 2600 -Min 1800 - Max avg 2400)
BUT I now have an alternative, if CCE decides to go on strike again.
So, it's definitely good to have it there, as plan B.
Thank you dvd2svcd!

[DDogg]

@24hourloop
What was your min bitrate set at?
What filter? If bicubic, what values?
What motion search precision?

Did you try CQ?

Without these values being reported we are left in the dark.

@Arianos, you used 1800 as your minumum???

[Arianos]

@DDogg: Yes. I tried those settings some time ago, to see if the result would be ok, it was, and I had left them there since.
Here is the CCE excerpt from "Waterworld":
--------------------------------------------------------
- 23/2/2002 9:01:39 ìì
- Free on drive c: 27936.13 mb
- Free on drive e: 3779.30 mb
- Video Encoding
--------------------------------------------------------

Executing Cinema Craft Encoder.
DVD2SVCD no. frames: 194261
CCE no. frames: 194398
StreamSectors: 2511223764
AudioSectors: 197175132
VideoPAPO: 35845848
ScanOffsetBytes: 233278
SeqAligningBytes: 14826087
VideoEndHeader: 12
SubtitleSectors: 4810680
EmptySectors: 238.00
PictureSectors: 900.00
PureMPEGStream: 2258332726.93
Seconds: 8106.02
CDSize: 805.00
Cut point 797.00
Variable Settings:
Frames: 194398
Anti Noise Filter: 2
Passes: 5
Image Quality: 17
VAF file creation: On
Video Encoding Mode: Multipass VBR
Min. bitrate: 1800
Max. bitrate: 2600
Avg. Bitrate: 2228
Closing program
Video Encoding finished.
------------------------------------------

I guess I am pushing my luck? :)
P.S.
As I consider myself still a Computer Ignoramus (I just love movies...), did I do something stupid?

[Pko]

Originally posted by 24hourloop
CCE yielded an average quantitization factor (the smaller, the better) of 1.65 whereas TMPGenc only reached 3.83. Maximum quantitizations were equally distinct: 4.98 for CCE as opposed to 10.53. This does *not* mean that TMPGenc's picture quality is twice as bad as CCE's, it is merely a scientific way of looking at the compression quality and 'smaller is better'.


Well, first, I think that CCESP is a better encoder; is faster and I really like the streams it produces...

But I think that your comparation has not really a meaning (well, is an indicator, at least) because, if I am not mistaken, the Q factor simple means that the encoder had to discard some information because it did not has bitrate enough (or something like that). So, in the same conditions, an encoder with less Q means that it has been more efficient in his compressing, and so is "better" in some sense.

But I think this has its flaws; first, we are making a lossy and *perceptual* compression; we are discarding some information of the oreiginal picture because we think that it will not be perceived (or be "less perceived" than other information) by the viewer. So, the Q factor DOES NOT measure that; perhaps the encoder has done it so, although the Q factor is high, the appearance to the viewer is better than some other compressor with less Qf

Also, I said "in the same conditions", but perhaps CCESP applys some filter before encoding that makes it mucho more compressible, although since it modifies the image, some information must be lost; I try to show it with an absurd example: suppose you aply a "filter" to the stream; that "filter" simply erases EVERYTHING, each frame is blank. So it will compress perfect, and the Qf will be 0. The image has a perfect compression, but the image compressed is not the original, but something much more compressible.

In fact, I think that CCESP uses some filters that "softens" or something the image; the result is usually very good and compresses well and fast, but the result is less "truer" to the original... In some cases like "saving private ryan", for example, that "softing" really kills one effect, the "roughness" of the film, that the original creator intended to be in the movie...

I think the CCESP results are better than TMPGEnc and I use it, but precisely because they are SO MUCH better and also is so much faster, I suspect it uses some kind of trick... perhaps someday TMPGEnc author discovers something like that or someone makes a avisynth filter that let TMGEnc get results more on par...

Well, perhaps is all simply better coding and I am dead wrong, but that is what I think right now...

[friar01]

i've been encoding for only about 6 months now, using tmpgenc. after many conversions, i began to get impatient with how long it took tmpgenc to convert a file (somtimes taking almost a whole day with 2 pass vbr). so i began to look for other encoders. i came upon cce a few days ago, and fell in love with it immediately. the quality, i think, is definitely better than tmpgenc's, and much faster (on my p4, it takes just as long for cce to do 5 or 6 passes as it does for tmpgenc to do 2 passes). there is less pixelation in higher action scenes; everything just looks absolutely amazing. i don't think i'll ever go back to tmpg for encoding. of course, there is the little problem with the watermark, but since i "bought" :) cce, that's been taken care of as well.

[24hourloop]

Originally posted by DDogg
@24hourloop
What was your min bitrate set at?
What filter? If bicubic, what values?
What motion search precision?

Did you try CQ?

Without these values being reported we are left in the dark.

@Arianos, you used 1800 as your minumum???


I used 300 as min, 2520 as max. Simple resize. Motion search precision high. No CQ.

[24hourloop]

Originally posted by Pko
Well, first, I think that CCESP is a better encoder; is faster and I really like the streams it produces...

Of course you are correct in your statements. It is very hard to make any sort of qualitative statement about something as subjective as picture quality. I tried very hard to relativize my statement but apparently failed to make it clear enough.

I like CCE and I like TMPGenc just as much in terms of what I can personally perceive on a 33" TV (or the PC monitor, which is much more critical).

CinemaCraft boasts that their speed is due to the fact that their core routines are handcrafted in assembler, which would be a believable explanation.

Also, of course both encoders apply some preprocessing. That also is extremely hard to judge by the numbers.

I am not sure your chracterization of quanitization is right on the money, though. The heart of lossy compression is the quantization step in which the transformed coefficients (which can be considered as continuous floating point numbers) are converted into representations using smaller numbers of bits and therefore are represented with larger "quantum" steps. The bigger the quantitization the wider the range of frequency being reduced.

Of course, if you start with a lower frequency (a smoothed image) you have to compress less. However, if you start with a higher frequency and have to compress more the reverse process will yield (up to a point) worse results than if you had started with a smoothed image (pixel noise).

Thus your statement that since CCE preprocesses (more) and has because of that a lower quantitization still means chances are it produces a better image.

ON the other hand I dare anybody to watch both movies and to tell me which was encoded how. I sincerely doubt that any of us mortals can see the difference.

[Pko]

Originally posted by 24hourloop

ON the other hand I dare anybody to watch both movies and to tell me which was encoded how. I sincerely doubt that any of us mortals can see the difference.

That is the point, IMHO...

When you cannot distinguish the original from the compressed, the encoder is doing a perfect work (for you as a viewer, since perhaps other people can make one from the other). That is, you are show both streams and you do not know what is original and what is compressed and you cannot tell by just watching.

Unfortunately, that usually does not happen... but that is a pretty objective measuring (it depends only in the viewer's hability, not in the viewer's taste)

When you CAN tell the original from both compressed streams, then you can use some kind of mathematical measurement, that will be objective but that will not be really useful if you cannot control ALL the process, to isolate only the COMPRESS step, or you can use a totally subjective measurement, that, odd enough, is what really matters, since the people could even say that they prefer the compressed stream to the original! that has happened once to me, when the more "softer" look of the CCESP compressed (and avisynth filtered, deinterlaced, etc.) stream look actually nicer for my personal taste than the grainy original!

Because of that, I think that numerical comparisons really do not matter: you must compare results by the naked eye and then select your winner... The math can be useful to understand other things (like why some streams look better than others using the same settings and encoders), but I think it is not so much useful to compare different encoders.