This from Terry on the MUG list;
================================
Here are some quality tests I did over the last few days comparing MSPro6 with the Ulead DVD Plugin to TMPGEnc.
Remember: all MSPro MPEG exports done with the DVD Plugin in place because it installs an updated version of the Ligos GoMotion codec.
One way to quantify MPEG quality is to measure its "Q". The Q is a measure of temporal quantization, or compression. Generally speaking; the higher the Q, the lower the quality, with Q having a maximum value of 31. I measure Q using the BitRateView program, which you can get here;
http://tecoltd.com/bitratev.htm
Now....Q isn't the only objective measure of MPEG quality. There are other things like peak signal-to-noise ratio and the like that require stream analysis software to do the measurements (not cheap), but Q is the easiest (and cheapest) for most users to measure. It also sticks to the basic practice of using the degree of compression as an indicator of potential quality. So....Q it is for the purposes of these tests.
Just like with MJPeg and other compressors artifacting and subjective look have to be determined by eye.
That said....here we go....
============== SETUPS ============
Systems (both): Asus P3B-F, PIII/850, 512 megs
Sources:
Both the YUY2/HuffYUV and DV sources were recorded from the same 4002 frame sequence of kids rampaging in a playground (VERY high action sequences).
YUY2/HuffYUV clips captured by AVI_IO 3.19 using a Matrox Marvel eTV & S-Video to 352x240 (VCD) and 480x480 (SVCD).
DV clips caputred by MSPro6's Video Capture using the RT-2000's IEEE-1394 device to 720x480
=========== PARAMETERS ===========
VCD settings:
The encoders preset, highest Motion Compensation (20 for MSPro6, "Highest Quality" motion search for TMPGEnc)
SVCD settings:
NTSC CBR
Frame Size: 480 x 480, 29.97 fps
Video Data Rate: 2520 kbps (TMPGEnc's SVCD CBR max, to keep it even)
Audio Bit Rate : 192 kbps
Motion Compensation = highest setting
Time = rendering time expressed as MM:SS
=========== VCD/SVCD CBR ==========
Encoding YUY2/HuffYUV source file:
MSPro6;
VCD: Peak 12.41 Avg 09.01 Time: 04:14
SVCD: Peak 15.08 Avg 09.68 Time: 09:40
TMPGEnc 12f;
VCD: Peak 08.54 Avg 06.11 Time: 12:06
SVCD: Peak 18.24 Avg 12.88 Time: 32:45
Encoding DV source file:
MSPro6;
VCD: Peak 10.18 Avg 07.35 Time: 05:53
SVCD: Peak 15.48 Avg 10.81 Time: 12:05
TMPGEnc 12f;
VCD: Peak 08.14 Avg 05.29 Time: 23:08
SVCD: Peak 18.24 Avg 11.43 Time: 48.21
======
Looks to me no matter what format the source file is, when encoding with CBR you're likely better off with TMPGEnc for VCD and MSPro6 w/DVD Plugin for SVCD. The DVD Plugin is also faster.
============= SVCD VBR ============
Now, lets see what happens if we encode with VBR for SVCD. The SVCD VBR setttings are the same as with CBR with the following exceptions;
MSPro6 W/DVD Plugin: Program default, but with 2520 kbps to match TMPGEnc's max.
TMPGEnc 12f: Manual VBR, 2520 kbps peak, 0 minimum, highest quality Motion Search
======
Encoding YUY2/HuffYUV source file:
MSPro6;
SVCD VBR: Peak 14.32 Avg: 10.04 Time: 09:47
TMPGEnc 12f;
SVCD MVBR: Peak 20.69 Avg 12.63 Time: 36:05
Encoding DV source file:
MSPro6;
SVCD VBR: Peak 15.95 Avg 10.92 Time: 14:22
TMPGEnc 12f;
SVCD MVBR: Peak 21.42 Avg 14.21 Time: 39:12
======
MSPro6 w/DVD Plugin is doing pretty good with SVCD, 'eh?
========== DVD:CBR/VBR ==========
Now let's see what happens when we generate 5500 kbps DVD files using the same DV source;
(note: TMPGEnc 2-pass VBR was done at 5500 kbps max/4000 kbps avg.)
MSPro6;
DVD CBR: Peak 10.94 Avg 06.59 Time: 18:06
DVD VBR: Peak 11.48 Avg 06.96 Time: 16:42
MPGEnc 12f:
DVD CBR: Peak 14.11 Avg 09.71 Time: 60:40
DVD VBR: Peak 14.15 Avg 09.80 Time: 64.39
DVD 2passVBR: Peak 17.97 Avg 12.77 Time 120:40
======
Generally speaking TMPGEnc has the advantage for MPEG-1/VCD while MSPro6 w/DVD Plugin has one for MPEG-2 formats like SVCD and DVD. This is especially the case given that either Ulead or Ligos has chosen to limit the updated GoMotion MPEG-1 engine to just 2000 kbps instead of its previous capability of 14,648 kbps. So much for xVCD.
One factor complicating this simplistic logic is that TMPGEnc 12f VBR files run about 15% smaller & 2-pass VBR runs about 35% smaller. . On the other hand MSPro6 w/DVD Plugin is a LOT faster and generally will deliver a lower Q. Pick your priorities.
Note: the lowest Q seen at any single point in a file during these tests was in the MSPro6/DVD Plugin DVD VBR flie: 3.41
========= VBR bitrate overruns =========
Much has been made of bitrate overruns in both TMPGEnc and Ligos GoMotion when using VBR. This could cause real problems if the peak bitrate exceeded the capability of the deck to display it. Even if it were momentary it could throw off the player for the remainder of the playback session.
Fortunately many, if not most, VCD/SVCD compatable decks can also support xVCD and xSVCD bitrates. However, there are still plenty that do NOT, and these are the players that a bitrate overrun will trip up.
In these tests Ligos GoMotion peak VBR bitrate exceeded the settings by from 559 (DVD) to 1019 kbps (SVCD).
There were NO bitrate overruns with TMPGEnc using either MVBR @ 5500 kbps or 2-pass VBR @ 5500 max/4000 avg kbps .
What is interesting is the timing of this bitrate overrun in Ligos GoMotion: the BitRateView graphs clearly show that this bitrate overrun ALWAYS occurs at the very beginning of the encoded file, and then it drops to below the set value within the first 1 to 3 seconds. The bitrate then sticks to its maximum bitrate setting or below for the remainder of the files duration.
So....it seems that the real problem is the Ligos GoMotion encoder takes a second or so to settle down when it first starts.
I tried everything from adding a black color clip to doing a 2-3 second fadein at the beginning of the project in an attempt to flatten this spike, but to no avail with 2520 kbps SVCD destined clips. With the 5500 kbps DVD files, however, a 2 second fadin got rid of it. This may or may not be the case with all DVD VBR files.
Perhaps the best solution, until VBR gets fixed, is to simply add a few extra seconds to each clip before encoding and then cut out this portion of the MPEG before feeding it to the authoring software. I have found TMPGEnc's MPEG Tools to work quite well for trimming this high bitrate portion off of MSPro6 MPEG VBR files.
Now the next step: send a note to Ulead about this....
=================================
Dr. Mordrid
[This message has been edited by Dr Mordrid (edited 30 May 2001).]
================================
Here are some quality tests I did over the last few days comparing MSPro6 with the Ulead DVD Plugin to TMPGEnc.
Remember: all MSPro MPEG exports done with the DVD Plugin in place because it installs an updated version of the Ligos GoMotion codec.
One way to quantify MPEG quality is to measure its "Q". The Q is a measure of temporal quantization, or compression. Generally speaking; the higher the Q, the lower the quality, with Q having a maximum value of 31. I measure Q using the BitRateView program, which you can get here;
http://tecoltd.com/bitratev.htm
Now....Q isn't the only objective measure of MPEG quality. There are other things like peak signal-to-noise ratio and the like that require stream analysis software to do the measurements (not cheap), but Q is the easiest (and cheapest) for most users to measure. It also sticks to the basic practice of using the degree of compression as an indicator of potential quality. So....Q it is for the purposes of these tests.
Just like with MJPeg and other compressors artifacting and subjective look have to be determined by eye.
That said....here we go....
============== SETUPS ============
Systems (both): Asus P3B-F, PIII/850, 512 megs
Sources:
Both the YUY2/HuffYUV and DV sources were recorded from the same 4002 frame sequence of kids rampaging in a playground (VERY high action sequences).
YUY2/HuffYUV clips captured by AVI_IO 3.19 using a Matrox Marvel eTV & S-Video to 352x240 (VCD) and 480x480 (SVCD).
DV clips caputred by MSPro6's Video Capture using the RT-2000's IEEE-1394 device to 720x480
=========== PARAMETERS ===========
VCD settings:
The encoders preset, highest Motion Compensation (20 for MSPro6, "Highest Quality" motion search for TMPGEnc)
SVCD settings:
NTSC CBR
Frame Size: 480 x 480, 29.97 fps
Video Data Rate: 2520 kbps (TMPGEnc's SVCD CBR max, to keep it even)
Audio Bit Rate : 192 kbps
Motion Compensation = highest setting
Time = rendering time expressed as MM:SS
=========== VCD/SVCD CBR ==========
Encoding YUY2/HuffYUV source file:
MSPro6;
VCD: Peak 12.41 Avg 09.01 Time: 04:14
SVCD: Peak 15.08 Avg 09.68 Time: 09:40
TMPGEnc 12f;
VCD: Peak 08.54 Avg 06.11 Time: 12:06
SVCD: Peak 18.24 Avg 12.88 Time: 32:45
Encoding DV source file:
MSPro6;
VCD: Peak 10.18 Avg 07.35 Time: 05:53
SVCD: Peak 15.48 Avg 10.81 Time: 12:05
TMPGEnc 12f;
VCD: Peak 08.14 Avg 05.29 Time: 23:08
SVCD: Peak 18.24 Avg 11.43 Time: 48.21
======
Looks to me no matter what format the source file is, when encoding with CBR you're likely better off with TMPGEnc for VCD and MSPro6 w/DVD Plugin for SVCD. The DVD Plugin is also faster.
============= SVCD VBR ============
Now, lets see what happens if we encode with VBR for SVCD. The SVCD VBR setttings are the same as with CBR with the following exceptions;
MSPro6 W/DVD Plugin: Program default, but with 2520 kbps to match TMPGEnc's max.
TMPGEnc 12f: Manual VBR, 2520 kbps peak, 0 minimum, highest quality Motion Search
======
Encoding YUY2/HuffYUV source file:
MSPro6;
SVCD VBR: Peak 14.32 Avg: 10.04 Time: 09:47
TMPGEnc 12f;
SVCD MVBR: Peak 20.69 Avg 12.63 Time: 36:05
Encoding DV source file:
MSPro6;
SVCD VBR: Peak 15.95 Avg 10.92 Time: 14:22
TMPGEnc 12f;
SVCD MVBR: Peak 21.42 Avg 14.21 Time: 39:12
======
MSPro6 w/DVD Plugin is doing pretty good with SVCD, 'eh?
========== DVD:CBR/VBR ==========
Now let's see what happens when we generate 5500 kbps DVD files using the same DV source;
(note: TMPGEnc 2-pass VBR was done at 5500 kbps max/4000 kbps avg.)
MSPro6;
DVD CBR: Peak 10.94 Avg 06.59 Time: 18:06
DVD VBR: Peak 11.48 Avg 06.96 Time: 16:42
MPGEnc 12f:
DVD CBR: Peak 14.11 Avg 09.71 Time: 60:40
DVD VBR: Peak 14.15 Avg 09.80 Time: 64.39
DVD 2passVBR: Peak 17.97 Avg 12.77 Time 120:40
======
Generally speaking TMPGEnc has the advantage for MPEG-1/VCD while MSPro6 w/DVD Plugin has one for MPEG-2 formats like SVCD and DVD. This is especially the case given that either Ulead or Ligos has chosen to limit the updated GoMotion MPEG-1 engine to just 2000 kbps instead of its previous capability of 14,648 kbps. So much for xVCD.
One factor complicating this simplistic logic is that TMPGEnc 12f VBR files run about 15% smaller & 2-pass VBR runs about 35% smaller. . On the other hand MSPro6 w/DVD Plugin is a LOT faster and generally will deliver a lower Q. Pick your priorities.
Note: the lowest Q seen at any single point in a file during these tests was in the MSPro6/DVD Plugin DVD VBR flie: 3.41
========= VBR bitrate overruns =========
Much has been made of bitrate overruns in both TMPGEnc and Ligos GoMotion when using VBR. This could cause real problems if the peak bitrate exceeded the capability of the deck to display it. Even if it were momentary it could throw off the player for the remainder of the playback session.
Fortunately many, if not most, VCD/SVCD compatable decks can also support xVCD and xSVCD bitrates. However, there are still plenty that do NOT, and these are the players that a bitrate overrun will trip up.
In these tests Ligos GoMotion peak VBR bitrate exceeded the settings by from 559 (DVD) to 1019 kbps (SVCD).
There were NO bitrate overruns with TMPGEnc using either MVBR @ 5500 kbps or 2-pass VBR @ 5500 max/4000 avg kbps .
What is interesting is the timing of this bitrate overrun in Ligos GoMotion: the BitRateView graphs clearly show that this bitrate overrun ALWAYS occurs at the very beginning of the encoded file, and then it drops to below the set value within the first 1 to 3 seconds. The bitrate then sticks to its maximum bitrate setting or below for the remainder of the files duration.
So....it seems that the real problem is the Ligos GoMotion encoder takes a second or so to settle down when it first starts.
I tried everything from adding a black color clip to doing a 2-3 second fadein at the beginning of the project in an attempt to flatten this spike, but to no avail with 2520 kbps SVCD destined clips. With the 5500 kbps DVD files, however, a 2 second fadin got rid of it. This may or may not be the case with all DVD VBR files.
Perhaps the best solution, until VBR gets fixed, is to simply add a few extra seconds to each clip before encoding and then cut out this portion of the MPEG before feeding it to the authoring software. I have found TMPGEnc's MPEG Tools to work quite well for trimming this high bitrate portion off of MSPro6 MPEG VBR files.
Now the next step: send a note to Ulead about this....
=================================
Dr. Mordrid
[This message has been edited by Dr Mordrid (edited 30 May 2001).]
Comment