Some people were asking about the speed and quality of different mainstream mpeg encoders - so i put together a little test.
SOURCE = DV 6230 FRAMES , 03:27:20
FAST PANS/SCANS AND FIRE EFFECT SOURCE MATERIAL OFF A D-BETA TAPE CAPTURED TO DV
ALL ENCODING WAS DONE AT 8000 MAX , 6000 AVG , 1500 MIN , 9 BIT DC
ALL SHOTS BELOW ARE FROM FRAME 1006 ...
NO AUDIO ENCODING WAS DONE!
ALL WERE ENCODED ON A DUAL XEON 2.4 GHZ WITH 2GIG MEMORY. SOURCE AND TARGET DISCS WERE ULTRA320 SCSI DISKS (DIFFERENT DRIVES FOR SOURCE AND TARGET) , AND NOTHING ELSE WAS RUNNING.
CCE SP 2.50
2 PASS VBR (PLUS 1 PASS VAF = 3 PASSES) : 4:49
CCE SP 2.67
1 PASS VBR (PLUS 1 PASS VAF = 2 PASSES) : 3:14
2 PASS VBR (PLUS 1 PASS VAF = 3 PASSES) : 5:01
notes: set to natural picture easy with 4 dithering and 28 Flat part (other settings were OFF)
MAIN CONCEPT 1.4
VERY HIGH QUALITY STANDARD SETTING (WITHOUT FURTHER TWEAKING)
VCSD GOP - ON
2 PASS VBR : 13:23
HIGH QUALITY STANDARD SETTING
FAST GOP - ON
2 PASS VBR : 7:54
FAST GOP - ON
1 PASS VBR : 4:40
2 PASS WITH FAST MOTION SEARCH AND OPEN GOP: 6:14
SONY VEGAS 5.0
2 PASS , 31 SEARCH RANGE (BEST SETTING) : 8:07
CANOPUS PROCODER 2
2 PASS HIGHEST QUALITY SETTING 5:19
2 PASS MASTERING SETTING 5:27
CANOPUS PROCODER 1.5
2 PASS MASTERING MODE 9:12
NOTE: CREATED A WAV FILE ALSO - SO TIME MAY NOT BE TAKEN AS A DIRECT COMPARISON
CCE 2.5 2 PASS
CCE 2.67 1 PASS
CCE 2.67 2 PASS
MAIN CONCEPT 2 PASS HIGH QUALITY
TMPGENC 2 PASS
SONY VEGAS 5
PROCODER 2 - HIGHEST QUALITY
PROCODER 2 - MASTERING QUALITY
PROCODER 1.5 MASTERING QUALITY
SAME FRAME FROM SOURCE DV FILE
+ Reply to Thread
Results 1 to 30 of 253
Thread: MPEG2 Encoder test
"Each problem that I solved became a rule which served afterwards to solve other problems." - Rene Descartes (1596-1650)
the above frame sample is a very tough part of the video file to encode.... al the sample shots are taken with virtualdubmod and all were shown as B frames ..
here is just a few frames of that sequence in a small dv file if you want to run your own tests.
right click and save as (4.5 meg compressed - rar file):
http://tinyurl.com/4sdxb"Each problem that I solved became a rule which served afterwards to solve other problems." - Rene Descartes (1596-1650)
Hmm...So apparently MainConcept is the worst?
mistake post"Each problem that I solved became a rule which served afterwards to solve other problems." - Rene Descartes (1596-1650)
Procoder does not look "exceptional". High quailty 2-pass shows some banding in the top rgiht hand corner. Mastering quality 2-pass is blurrier than all the others! CCE and TMpeg do not show banding in that same are during 2-pass. And yes, Mainconcept is fugly. I've got TMPEG but have practically bent over backwards trying to get it to work. I guess I will be trying some more. I was going to go with the Mainconcept encoder! Eyuck!Your miserable life is not worth the reversal of a Custer decision.
keep in mind that 1 frame does totally indicate the quality of the whole sample .. i also compressed the images to jpeg (but i compared to the png source from v-dub and they look pretty close the same)
also - these are B frame shots .. the closest I frame to this in all cases was better looking
if requested - i can post the same I frame of each same seq.
Originally Posted by BJ_MYour miserable life is not worth the reversal of a Custer decision.
No big surprise about Mainconcept, i have also noticed during testing that TMPGEnc is much better. The big surprise for me was that procoder could do that bad, i never tried it but from what i read here many forum users say its so good. Also a bit surprising that Tmpgenc could do that much better than CCE, i would have expected those two to be very close, and i think they could have been closer if minimum was set higher. Its just a guess anyway...
Would be nice if you posted I-frames also, or just tell us how you would rate them.
Originally Posted by Sillyname
I've never seen with a naked eye blocks coming out of ProCoder even on the low bitrates, but unfortunately with version 2 it got changed. Therefore I'll be sticking with ProCoder 1.5, which for DV PAL material is truely unbeatable. And remember that the flaws are easier spotted in PAL because it's got better everything than NTSC: better resolution, better colors, better contrast ratio, etc., so all the MPEG2 crap will be seen there for sure, while with NTSC various MPEG2 encoders will seem to have less vatiation in quality. At least to your naked eye.
So for me the only encoder that gives satisfactory encodes from DV PAL is ProCoder 1.5. I recommend it instead of ProCoder 2 or any other commonly available MPEG2 encoder. I tried CCE SP on several occasions for DV PAL and it looks like crap. While trying to use its filters may help slightly, unfortunately they make the motion look more progressive/film-like which is no good for interlaced look results. TMPGEnc and MainConcept are not much better either - each one has certain shortcomings in image quality. However, I really like TMPGEnc's noise reduction capabilities, so sometimes I use it for AVI->AVI noise reduction (for VHS mostly) before doing the MPEG2 encoding in ProCoder. And on a side note, I love MainConcept DV Codec.
P.S.: For progressive footage CCE SP is quite good, but for DV I think it's hard to beat ProCoder 1.5.
There must be something wrong, either CCE is the only encoder making good P-frames or its not the same frame. Look at the top right, that cant be the same frame.
cce was 1 frame back compared to the others ...
like you said they all look better than the B-frames, but Main Concept still looks the worse to me. I never could get it to work correctly on my system anyway so now I don't feel like I'm missing so much. I'll stick with TMPGEnc Plus 2.5 for a while yet.
MC looks blurry, even compared to TMPGENC. No surprise.
TMPGEnc looks decent. No surprise. I would have done ESTIMATE instead of FAST (not much slower, and tends to look cleaner with less block artifacts).
CCE looks like a grainy EP VHS tape. No surprise. It did not inherently clean the video like MPEG does, and it added some noise of its' own. Terrible encoder, don't know what people see in it.
But this really gets me ... why does HIGHEST seem to look better than MASTERING in Procoder in the first example? Strange. Maybe it's just the footage, or maybe you got them backwards?
Originally Posted by lordsmurf
You oughtta subtract 2 of them and see what happens.
Smurf, CCE terrible? Are you trying to say that Procoder won this test? I think CCE looks better than Procoder in this test, the B frames have very obvious differences and the P frames are close but cant say for sure since its not the same frame. I agree that CCE got more noise, but in my opinion, with the bigger difference in B frames quality, Procoder comes out below CCE in this test. So if you can say CCE is terrible, what would you say about procoder? Cant be a word you teach to kids
Maybe i will test a bit myself later this weekend, for me it looks like CCE has considered frame 1006 to be a place to save bitrate, while tmpgenc has used "too high" bitrate. Since some people say tmpgenc seems to be afraid of going too low when doing VBR i find reason to believe that the tmpgenc sample is closer to a CBR encode. However i do not question tmpgencs quality at high bitrates, it is possible the best in that area, so it probably is as good as this test shows.
Originally Posted by FOO
thor, absolutely Procoder won this one. Just look at the noise in the CCE vs the Procoder. I want clean video.
I acknowledge that the DV source was noisy, and that CCE may have been the truest conversion of them all, but it translated the noise AND added some of its own. That's not acceptable.
In this example, I think TMPGEnc did even better than CCE.
funny how vegas has more detail than the main concept stand alone (since they both use the main concept encoder).... sony has tweaked it they say ... vegas was the only one that picked out the small double line in the middle right lower third in frame 1006
-- it may have some other problems , but that caught my eye
Hi all. TGIF :P
..Bout time someone finally did a head-to-head MPEG test.
I think it would have ben better if WE all could have part taken
in the encoding too.
It would be nice if there were such a test scenario to partake
in w/ our DV cams on our own, but that would probably be too
difficult to accomplish - sigh. Anyways.
In my OP, I thought that TMPG proved itself very well there.
My thought on Interlace and CCE..
I'm not too sure that there is really any serious issues w/ CCE
encoding Interlaced source. Maybe its just a matter of properly
orienting the source's video Aspect Ratios ??
I'll have to test this out. Perhaps over this weekend. I have
CCE v2.50 I think. Every time I think I don't need it, I end up
digging it out again for some domb testing :P ..I guess that's
why this Hobby is soo much fun Anyways.
I think that for Interlaced (ie, DV) CBR would be best. I think
that at least for when the source was has ben shot by Hand, vs. by
a Tripod. If footage was doe w/ a Tripod (fluid would be best) then
there is no limit to what Encoding Mode one chooses in their encoding.
I still wish there was some sort of "template" that everyone w/ a DV
cam can use to shoot footage at and then test encoding scenarios against.
That would be most interesting. Any ideas anyone ?? Anyways.
Thanks for your interesting thread.
Question. How does one obtain the B frames pic's ??
Ok, but i think CCE kept more detail, so it was number 2 for my eyes. Tmpgenc got the first place. In some areas i see less noise with Procoder, but i also see more blocks. Makes sense after reading Edmund Blackadders post. Maybe Procoder 1.5 and CCE 2.50 should have been included in this test. Anyways, big thanx to BJ_M.
ADDED procoder 1.5 and CCE 2.5 above
I see blocks in PC 2.0 but not in PC 1.5
Are you sure the upgrade was really better?
Same for CCE. I think 2.5 is doing better than 2.67.
Originally Posted by lordsmurf
CCE 2.5 looks better (same settings were used) and was faster - on this source ..
PC 1.5 was slower and looks better - i agree , on this source ...
PC2 has better scene detection, filters, settings, and a host of other features ...
i re-ran the mastering PC2 test again with stock settings and CBR 7500 and it looks better than 2 pass VBR , but not fair test as it is a higher avg bit rate ..
this is a brutal sample i used for testing - it was not used to "look pretty" but to cause a problem .. you can download a small sample of it above and try to encode it any way you want ..
some filters will help - but lose detail .. that fire/explosion is over in 1 second apx.
i didnt use any real super tweaks or any filters -- tmpgenc, MC , vegas, PC and to a degree , cce - all have them ...
PC looks suprisesly bad and TMPGenc suprisesly good to excellent!
CCE looks as it is expected (I read everywhere that 2.5 is better later version, but I don't test that myself) and MC blured as expected (terrible, bad, awfull!).
Nice work BJ_M! Thanks!
Just for the info, I'm still using TMPGenc plus for encoding offline my projects and I use MC only for realtime mpeg 2 captures from TV / non important VHS tapes. And after this test, I'll keep doing that!La Linea by Osvaldo Cavandoli
Good work BJ_M!
The MC B-frame sample shows the major problem with MC. Even with two pass VBR, MC will rarely pump-up the bitrate on demanding scenes causing bitrate starvation effects in high motion frames. The blocky appearance indicates too much pixel truncation during encoding of the frame.
I believe Tmpgenc has a similar problem with low motion scenes. Contrary to MC that doesn't deviate much from the average bitrate, Tmpgenc is able to do so, and is aggresive in low content dark scenes, causing the image to become flat black in large areas. This is clearly visible in high contrast capable displays (flat panels mainly).
CCE has a mind of it's own. It has a BIAS parameter that apparently controls the bitrate fluctuation but in low bitrates it tends to soften the picture.
It appears that MPEG encoding implementations have a long way to go before they can produce excellent results.
BTW, a good tool to evaluate the encoding is to use DVD-LAB's bitrate viewer on the MPEG files. It clearly shows how bitrate is distributed along the time axis and how aggressive the encoder is to use bitrate when it is needed.The more I learn, the more I come to realize how little it is I know.
"What? Huh?!? WHAT will come out no more?!?" Jack Burton -- BIG TROUBLE IN LITTLE CHINA
Might I suggest a new test?
Try filming a strobe light at it's fastest setting. The adjacent frames that do not contain a flash should whig-out when the mpeg encoder tries to compare with a frame containing the flash. I found the Mainconcept encoder to be virtually worthless when compressing this sort of scene. TMPEG rips right through it, though! Still wouldn't mind seeing some B-frame comparisons...
What happened to my avitar?Your miserable life is not worth the reversal of a Custer decision.