Why is --qcomp 0.6 default for 2passes?

[Audionut]

Originally Posted by Trahald

Once a thread becomes 1 or 2 people beating a dead horse I generally tend to close them, Im not sure we are there yet although we are close.

Imo, there has been more than enough explanations by more than enough people answering the op's original questions.

Originally Posted by simps

Can someone explain why --qcomp 0.6 is default for 2passes, what is the logic behind that?

Vote 1 for thread closure.

[simps]

I am sorry I am logicaly minded.
When I post a scenario where a higher qcomp up to 0.89 is a overall WIN vs lower qcomps, it is obvious too assume, that any lower value than 0.89 would also win vs even lower values.

In other words, If 0.89 is better overall than 0.7, it is obvious that it would also be better than 0.6, or even 0.8 would be better than 0.6 too, and 0.75 would also be better, so you got the point.

You are showing a scenario where a 0.89 qcomp is not better than a 0.6 qcomp. In this case, YOU CAN'T SAY ANYTHING ABOUT HOW A 0.7 OR 0.75 QCOMP WOULD DO AGAINST 0.6. Or even how a 0.65 qcomp would do. All those values are higher than default.

That is why I told you you need to try those outs too.
This is what I would expect from a logicaly minded person.

And the porpous of the thread is not "0.89 is the best", is to find a higher and better overall qcomp value.

And just to let you know, I have Rambo IV HD too, and will use it, same scenes, res, bitrate you did, to show how a 0.7 against 0.6 would do, or 0.8 vs 0.6, and so on.
you can post it yourself too, but obvious after this, I won't trust any post from you anymore.

Cheers,
Simps

[simps]

Also about closing the thread, not a good idea. There are some interesting info here and discussion too. Some times people like you come arround to trash talk the thread. This happens all the time, unfortunatly. For me, its a piece of cake to get pass this, don't close anything. Not sure for you though.

[Sagekilla]

There really is no new concepts in this thread at all though. qcomp controls how bit distribution occurs between scenes.

For bitrate (Non-CRF or QP) low values of qcomp cause it to allocate bits in a CBR-like manner where each frame gets a similar number of bits. As everyone knows, low motion scenes need fewer bits than high motion scenes because the residual is smaller when not a lot of motion is happening.

On the other hand, high qcomp values cause it to allocate bits unevenly. Instead of forcing each frame to have the same number of bits, it tries to give each frame similar quality.

Unfortunately, in bitrate mode this means high motion scenes tend to suck up inordinate amounts of bits and the overall quality of the whole movie suffers. Yes, your high motion scenes improve, but you won't be able to tell the difference if you're actually watching it!


I don't see why this thread has continued for as long as it has. It's simple, if you want higher quality in high motion scenes (at the expense of your low motion scene quality), you increase qcomp. The only reason why 0.6 is the default is because, like most other defaults, it provides a good balance of quality in your average movie watching experience. If you're really concerned with having nice quality high motion scenes, use CRF mode and increase qcomp slightly.

[simps]

Originally Posted by Sagekilla

There really is no new concepts in this thread at all though. qcomp controls how bit distribution occurs between scenes.

For bitrate (Non-CRF or QP) low values of qcomp cause it to allocate bits in a CBR-like manner where each frame gets a similar number of bits. As everyone knows, low motion scenes need fewer bits than high motion scenes because the residual is smaller when not a lot of motion is happening.

On the other hand, high qcomp values cause it to allocate bits unevenly. Instead of forcing each frame to have the same number of bits, it tries to give each frame similar quality.

Unfortunately, in bitrate mode this means high motion scenes tend to suck up inordinate amounts of bits and the overall quality of the whole movie suffers. Yes, your high motion scenes improve, but you won't be able to tell the difference if you're actually watching it!


I don't see why this thread has continued for as long as it has. It's simple, if you want higher quality in high motion scenes (at the expense of your low motion scene quality), you increase qcomp. The only reason why 0.6 is the default is because, like most other defaults, it provides a good balance of quality in your average movie watching experience. If you're really concerned with having nice quality high motion scenes, use CRF mode and increase qcomp slightly.

Agreed. I still want to post the mid and high bitrate comparison, to show you can get a overall better quality with higher qcomp. Is it going to be as high as in the low bitrate test? No. Maybe 0.8, 0.7, I don't know yet, need to do more tests.

Also, the idea of "adaptative qcomp" would be some BIG WIN, if possible. What do you think about that?

[Sagekilla]

Adaptive qcomp sounds absurd and unnecessary. I have reason to believe it would make rate control completely screwy and unreliable for that matter.

My point is that qcomp = 0.6 works fine for your average encode. In terms of crf, this means most people use values from 18 - 25 whereas your encodes of 240 kbps is akin to going to crf 30. As (I believe) Audionut said earlier, why ruin 95% of the encodes to benefit the 5% where higher qcomp matters? If you're -really- concerned with quality, you would find what works best for your source.

I mean, if you can find conclusive evidence on a wide variety of videos across different bitrates (low to high) and rate control modes (1-pass, 2-pass, crf, qp) that higher qcomp is more beneficial now to the overall quality, by all means post your results. But I don't think you're going to find any real difference at mid to high bitrates.

Remember, if you're seriously going to look at changing qcomp, do extensive testing on many different videos. One isn't going to cut it.

[kumi]

Originally Posted by Trahald

The audionuts screen shot of rambo close up looks better to me at default. more detail. but thats just mho.

Agreed.

[simps]

Originally Posted by kumi

Agreed.

But I agree to that too. This just shows --qcomp 0.89 is extreme for that mid bitrate and res. This doesn't say anything about how a --qcomp 0.7 or --qcomp 0.8 would perform, and this are both higher than the 0.6 default.

The only thing you can conclude so far is that for low bitrates with SD res, you can have better results up to some --qcomp 0.89 at least. There is evidence for this here.

The other thing you can conclude, is that for mid bitrates with 720p resolutions, a --qcomp 0.89 is too much and won't work for the best. And that is it, you can't say anything about other higher than default values, for example 0.7 or 0.8, for mid bitrates so far.

The problem here, is that people is kida slow minded (no offense to anyone). First when I stated the issue, people were thinking I was talking about --qcomp 1, which is not true. Now that I've posted that a --qcomp up to 0.89 is better for low bitrates at SD, people are using the 0.89 value to compare at 720p with mid bitrate.

The point I made, is that you CAN find a higher than default value of --qcomp for the bitrate you want. They will just not be a constant.

For low bitrates you might have some 0.89 optimal, for mid bitrates it might be some 0.72 and for high bitrates it could be 0.66. I am speculatin this numbers so far, just to get you an idea. But with my tests we will get to the correct numbers.

The point is, if you can find a higher --qcomp to suit you at any given bitrate, than why 0.6 is default?

This is the idea here, and what we are trying to find out.
It is too bad that people are slow minded and don't get the point, like Audionut doind that test with --qcomp 0.89, and concluding something that has NOTHING to do with what I am saying. That test does not prove me wrong.

Sorry for the "slow minded" term. My english is bad and I don't know what is the right term to use to define this.

For the mid bitrate discussion, you will need to wait for me to post my results. I will not do small clips like Audionut did, because that don't represent what the bit distribution will be when you encode the full movie. I will do it the right way, encoding full movie with each setting, and it takes time. I also have other things to do, so I won't be posting this today or tomorrow. When I have it posted here, than lets discuss the mid bitrate issue. And than, lets start to work on the high bitrate.

[Guest]

@simps

Since you're now resorting to accusing people of being slow-minded, and you are vacillating in any case about your claims, and your original question has been answered, it's time to close the thread.

If you eventually have some data to support a clear claim, then feel free to start a new thread to discuss it. But please refrain from personal attacks, per forum rule 4.