TMPGEnc Encoding @8000kbps CBR

[coeng73]

I just recently finished an hour-long video project using Premiere 6.5.

I then exported the file as AVI (I included the audio because my DVD authoring S/W cannot link audio and video files together).

I used TMPGEnc at 8000kbps CBR and produced an MPEG-2 in 9 hours time.

I burned the DVD successfully (with menus, submenus, menu music, the works).

I viewed the video on my 53-inch projection TV and was not completely pleased. Whenever there is the slightest motion of the camera, the video appears to be ever so slightly pixelated.

Is there any setting that I should have used in TMPGEnc to eliminate this?

I would like to get the DVD video to look as if I were playing it directly from the tape.

[Hitmandj]

TRY INCREASING THE ENCODING TIME.

This is done by increasing the option to higest quailty/ slowest.

[coeng73]

Hitmanj,

If you are referring to "Motion Search Precision" it is already set to "Highest Quality" (very slow).

Any other suggestions?

[coeng73]

I just did a side by side comparison (using Picture-in-Picture) of my DVD footage with my source footage via my camcorder (using S-video input) on my big-screen TV.

I queued up the same starting point and pressed play on both the DVD player and the camcorder simultaneously.

I notice two major differences.

The first difference is the issue that I originally posted: motion of the camera causes minor (but still noticable) pixelization.

The second difference is the difference in color. In my DVD the sky appears almost washed out and the outline of clouds are hard to distinguish. In the tape footage, the sky is a nice tint of blue and the clouds are well outlined as the scene is a Hawaiian sunset.

Any comments on this? I used my Pinnacle Studio DV card to capture the footage to DV, Premiere 6.5 to create the project, TMPGEnc to convert to MPEG2 at 8000kbps CBR, and DVD Workshop to author and burn the DVD.

[bfarnham]

Have you seen these settings. I am not an expert on this subject, but I have used these settings with very good results. Sometimes I use these same settings but with CQ instead of the 2-Pass VBR. A little less time and still good results.[/i]

[SaSi]

Don't use CBR encoding for high quality output. Use Constant quality, set it to 80 for good quality or 92 for excelent quality and set the max bitrate to 8000.

Then encode with motion search estimate or highest quality.

Why?

Because CBR produces a constant stream @ 8Mbps. If you have a low motion section followed by a high speed panning section (usual for home made camcorder videos), the bitrate is "saturated" and there is not enough bandwidth to hold the picture information.

By using Constant Quality compression, low motion parts don't saturate the bandwidth so there is plenty of reserve to boost bitrate for high motion parts.

This problem seldom shows in DVD source material (as cameramen and directors avoid such panning parts - to avoid making viewers dizzy).

Perhaps the bitrate calculator in Tmpgenc is "buggy" for CBR and works better in CQ mode.

[coeng73]

thanks for the info and suggestions. i'll be sure to post my results.

does anyone have an answer about the loss of color I'm experiencing?

[wway]

When I capture camcorder video using VirtualDub, I always have to adjust the brightness and contrast as described in the link below. Just a suggestion. Not sure if it applies in your case. Check it out.

http://www.lukesvideo.com/videotuning.html


wway

[coeng73]

In TMPGENc using CQ, the Max Bitrate can be set up to 100000 kpbs.
Why is this?

What theoritcally is the maximum bitrate and why?
The consensus seems to be that 8000 kbps is excellent.
What do DVDs that you buy and rent use?

What should the lowest value for minumum bit rate be set to for excellent quality?

Why padding or no padding? Is this only for the DVD authoring program?

I am looking to generate the best quality that I possible can. DVD space is not a problem. As long as I get the quality.

I also just tried an experiment where I set the max to 9000 and the min to 1000 without padding. I have yet to view it though. As soon as I do I'll post my results.

[coeng73]

Well I tried CQ using 9000 max, 1000 min, no padding.
Still get the same result.

Just the slightest camera movement results in noise. The noise almost looks like a mirage does in the desert or the hood of a running car in very hot weather.

I want to eliminate this at all costs.

[bugster]

coeng73, as far as your colour problems go, try setting 'output YUV data as Basic YCbCr' on the Quantize matrix tab of TmpGenc. I have seen something on these boards that said that was required fro DV material.

As for your pixelation, I would guess you are encoding home movie material. Unfortunatley, these are notoriously hard for any mpeg encoder to handle at any bitrate. Camera shake due to hand held cameras just eats up bitrate, so any or movement in the picture just makes it worse. You could try the 'soften block noise' option (again on the quantize matrix tab). This should reduce the visibility of the noise but at the expense of overall picture sharpness.

Hope this helps

[coeng73]

I already have YUV data as Basic YCbCr set.

I will try the other option, soften block noise.

[vh5150]

Use the noise reduction setting too. If you have the time though.

[coeng73]

I finally encoded my video to a level of quality that I am (mostly) satisfied with.

I used 2-Pass VBR, and used the "Soften Block Noise" option with "Intra-Block = 100" and "Non-Intra Block = 100". I don't know what these settings really mean (nor could I find much information on them). All I know is that the video produced SIGNIFICANTLY reduced the block noise I was experiencing.

To be honest, I do not see any "softening" of the picture as I read may occur when using the soften block noise option (especially since I used the max values of 100 as indicated above). Furthermore, my file size was only approx 2.5GB versus 3.9GB using CQ at 8000 kbps. (Better video quality & smaller file size: a win-win for me)

The only time I can really notice minor block noise is the few first frames of a chapter when it fades in from black video. I designed my project so that all my chapters are like this. I am willing to live with this momentary block noise, but was hoping if anyone knew why this happens. Once the fade-in to a chapter is done (each fade-in is 1 second in duration), the block noise virtually vanishes (until the chapter ends with a fade-out to black and the next chapter begins with another fade-in from black).

I hope this information helps anyone that experienced the same problem as I did.

Now I gotta work on that color problem that I'm having!

[SatStorm]

TMPGenc is not famous for encodings like this...
Better switch to ProCoder. It is better for such tasks

[wway]

I used 2-Pass VBR, and used the "Soften Block Noise" option with "Intra-Block = 100" and "Non-Intra Block = 100". I don't know what these settings really mean (nor could I find much information on them). All I know is that the video produced SIGNIFICANTLY reduced the block noise I was experiencing.

To be honest, I do not see any "softening" of the picture as I read may occur when using the soften block noise option (especially since I used the max values of 100 as indicated above). Furthermore, my file size was only approx 2.5GB versus 3.9GB using CQ at 8000 kbps. (Better video quality & smaller file size: a win-win for me)

The "soften block noise" option works in reverse. Where 100 = no softening and 0 = full softening. At least it works this way in TMPGEnc ver. 2.53. If you set it to zero, you get a blurry mess.

I don't know if it has been changed/fixed in later versions.

wway

[ronnylov]

My suggestion is to change the quantization matrices. Use a custom made matrice called "mb1 interlaced DV". I have tried it and it makes a big difference on DV sources.

Here's a link (on german, sorry):
http://www.dvd-svcd-forum.de/phorum/ikonboard.cgi?;act=ST;f=18;t=7499

Edit TMPGEnc.ini and replace the matrices section. Then choose 'mb1 interlaced DV' in the Quantize Matrize tab.

[SaSi]

The only time I can really notice minor block noise is the few first frames of a chapter when it fades in from black video. I designed my project so that all my chapters are like this. I am willing to live with this momentary block noise, but was hoping if anyone knew why this happens. Once the fade-in to a chapter is done (each fade-in is 1 second in duration), the block noise virtually vanishes (until the chapter ends with a fade-out to black and the next chapter begins with another fade-in from black).

Block noise in dark scenes appears to be a weakness in Tmpgenc. It might well be the matrices, however it only appears in relatively low bit-rates. Now, what is a low bitrate is an issue, but to give an example, if I compress an X-Files episode (a generally "dark" series) to fit on a CD, I get this effect. If compression is lower (to reach the 1Gb size each), blocky darkness seems to disappear or is reduced dramatically.

I suspect CCE is a bit better than Tmpgenc (on this and other issues). It is faster, early results show less artifacts and macroblocking, as well as higher image quality for a given constant bitrate. (better algorithms?).

However, I suggest you try Mainconcept's MPEG encoder (www.mainconcept.com. Early tests (2 days) have shown me that it is much much faster than tmpgenc (which I now find unacceptably slow for high quality compression), is possibly higher quality than CCE (a bit early to judge on that) - but certainly more versatile than both the above. A demo version allows you full use - save for a small watermark on the top left corner and a horizontal line that is usually green; it once became orange but no more I kinda preferred the orange one to the green one.

[jorbje]

coeng73: Your color/contrast problem is because you have switched "Output YUV data as Basic YCbCr not CCIR601" in tMPEG ON. I have also read that this should be ON if I encode DV-footage. But then I noticed the same problem as you have with ugly and hard contrast. When switching it back off, everything was fine.

[DJRumpy]

Sasi it correct, CCE doens't exhibit this problem. It's due to a fade. When you do a fade, each frame looks drastically different than the scene before, causing the encoder to wastefully throw bits, causing macroblocking when the bitrate is insufficient. Using multi-pass VBR should help with this, even in TMPGenc.