Constant Quality with MainConcept?

[cosmin]

Hi,
I've just downloaded the latest MainConcept MPEG Encoder trial (1.4.2), and I tried to encode a video using the Constant Quality setting (NB: not Constant Bitrate), without success. The CQ option is grayed (unavailable) all the time, no matter how much I fiddled with the settings.

I need the Constant Quality setting for a particular reason, and so far I was satisfied with what TMPGEnc is offering. But I did want to try MainConcept. How can I get it to work?

If I recall correcly, I tried Adobe Premiere 6.5 one year ago, and the CQ option was grayed, too.

Best regards,
Cosmin

[BJ_M]

enter a 0 (zero) in average and min and 8000 in max , then CQ is working ...

[cosmin]

Wow, how does that stand for an intuitive user interface .... NOT!

Thanks, BJ_M, for the tip. I also found it in the manual -- I was reading an older manual before, and this gimmick was not even mentioned.

This is probably off-topic, but in case anyone wants to know:
I didn't like that CQ has no upper bound on bitrate. That 8000 kbps (or whatever is) for the max value means nothing. On the other hand, TMPGEnc does a good job at limiting the quality below the user's option, if that's necessary to keep the stream below the 9800 kbps limit.

Best regards

[spectroelectro]

Since CQ is nothing but single-pass VBR, and since the MainConcept encoder is one of the faster encoders around, there's little point in using CQ as opposed to the standard VBR setting. I get slightly less than 1:1 (a little faster than real time) mpeg-2 encoding using typical VBR settings (min 2000, av 5000, max 8000) and perfectly adequate (read: small) encoded filesizes with the standard VBR settting in MainConcept.
If you were using TMPG, there'd be a big difference in encoding time twixt CQ and 2-pass VBR, but with MainConcept it's not an issue. So if I were you I'd just use the default 2-pass VBR setting. Your encoded file will be smaller and assuming you have a reasonably fast computer (> 1.5 Ghz) your encoding time shouldn't be an issue.

[cosmin]

You are absolutely right, spectroelectro - it's just that I had a different reason to use CQ: to make comparison tests. Here is the setup:

- Try to identify a way in which the CQ parameter matches the quantization scale. For example, I noticed that the CQ parameter in MainConcept is close to the actual value stored in the M2V stream, while there is a linear correspondence in the CQ factor in TMPGEnc and the actual value. In TMPGEnc, 60 corresponds to the factor 8; 75 corresponds to 5; 80 corresponds to 4; etc.

- Use bbtools to see the actual quantization scale. I also use a perl script that calculates the average (bbvinfo only shows the value for each frame).

- Use the same quantization tables across all encoders used in testing.

- Encode several videos at roughly the same quantization scale, and see which encoder achieves a smaller file size. Same quantization scale => same quality; smaller file size => better compression at the same quality.

Unfortunately, I couldn't reach to a conclusion yet, because I noticed that TMPGEnc cheats a little. Although I disabled all smoothing, it performs some smoothing on the P and B frames (which I can notice with VirtualDubMod), and this is how it wins the competition.

So far, I only tested TMPGEnc vs. MainConcept.

[BJ_M]

bitrate viewer will show you the avg Q levels -- or at any point

[cosmin]

bitrate viewer will show you the avg Q levels -- or at any point

Seems like a useful tool. Thanks for the tip!

[SaSi]

Unfortunately, I couldn't reach to a conclusion yet, because I noticed that TMPGEnc cheats a little. Although I disabled all smoothing, it performs some smoothing on the P and B frames (which I can notice with VirtualDubMod), and this is how it wins the competition.

I would not call this cheating. If the encoder "discovers" that the bitrate is not enough for a precice representation of the source video, it has to become proactive and find ways to minimize loss.

CCE will also smooth the images at low bitrates. MC tends to avoid doing that, generating blocky looking pictures at low bitrates.

Additionally, the selection of quantization coefficients can generate a soft looking image.

Comparing MC vs Tmpgenc, at low bitrates,
Tmpgenc tends to give a softer yet more pleasing picture quality and the weak spot is in dark and black scenes. The threshold from dark grey to total black is very low causing black areas on the screen without any picture information on them.
MC tends to generate more grainy pictures. In dark scenes it tends to genrerate a moire type effect on the screen by allowing too few colour values after the quantization.

In low bitrates, only CCE can perform and that at the cost of a really soft picture. At very very low bitrates (e.g. 2Mbps at 720x576) the image starts to become blurred, like the focus is off by a little.

At high bitrates (>7Mbps), differences tend to diminish, as there is really little to do.
]