This have been discussed before, right?
I encoded "The Day After Tomorrow" with OPV, 2pass and 10pass. And one underwater scene together with a thunderstorm looked awful with both OPV and 2pass. BUT... with 10pass the scene looked not far from the original. I figure that 10pass is a bit overkill in regard to passes so what I'm wondering is if anyone here could tell me how many passes would be enough?
Let's say for example that after 5 passes the quality isn't noticably improved.
Hit me!

__________________
AudioSlave

Hi audioslave,

Try 1-pass.
Then 2-pass.
Then 3-pass.
Then 4-pass.
Then 5-pass.
Then 6-pass.
Then 7-pass.
Then 8-pass.
Then 9-pass.
Then 10-pass.

I guess that's the only way to compare the files, and see which looks better, and after what pass there's no need to go further

-kwag

I knew someone was gonna say that! But, I thought it would be Dialhot. No offence...
Yes, Karl, you're probably right. I was afraid I was gonna have to do this...

__________________
AudioSlave

That's exactly what I thought when I was posting That's what I would do
Because that way you can compare against a constant source.

-kwag

Is there an easy way of taking just that one scene and do the test on that? I don't know how to set up CCE the correct way without using DVD-RB...

EDIT
I found out how to encode that scene and I'm now performing the test. I will let you know what I find out in a short while...

__________________
AudioSlave

An important observation: Multipass does not show a lot of improvement in a small piece, because it needs a resonable ammount of footage to "spread" bitrate and therefore make a better distributed bitrate curve (in theory).
Since CCE updates the VAF file on each pass, you can do one pass at a time and keep the files.

Alright.

The scene I've been experimenting with is the last 16 seconds of chapter 3 (From the DVD "The Day After Tomorrow):
A boy is being filmed from underwater and above the surface rages a thunderstorm with lots of lightning flashes.

The results of this test clearly shows that 10-pass (9+1) encodes is THE way to go. The 10-pass encode has more details and less artifacts compared to the other encodes. I know it's been said that 3 passes will do the job and everything above is a waste of time. But I know what I see and I know what I like so from now on I'm doing 10-pass encodes .
The only downside is that it will take me about 16 hours (!) to encode a 2 hour movie .

__________________
AudioSlave

@vmesquita
You're probably right, my friend. However, from this 16 second clip I really got all the proof I needed .

__________________
AudioSlave

Too short
Use some footage of at least 4 to 5 minutes

-kwag

@vmesquita & kwag
Alright, just for you guys . Maybe the whole chapter 3 would be a better idea? It's 2:53...

EDIT
Now re-ripping both chapter 3 and 4.
5:34 in length...

PS
I can use the same *.vaf for all encodes, right?

__________________
AudioSlave

3 passes give comparables results with tmpgenc, 4 passes for better results and 5 passes for best if you have patience and free time.

@jeo
Based on what? Your own testing or hearsay? Just curious...

__________________
AudioSlave

based in my tests and from many others forums tests.

And by 5 passes you mean 1 analyzing pass + 4 encoding passes, right?

__________________
AudioSlave

I tested this longer clip with both 5 and 10 passes (thanks for the tip jeo ) and there's not very much quality difference between them. In fact it looks as if the 5 pass clips have more details preserved! And only took half the time to encode .

__________________
AudioSlave

The biggest problem with multipass VBR in DVD-RB is that the whole movie is encoded in small parts. With multipass VBR it's very important that the saved bits would get divided across the whole movie, not just one cell.

One cell might be a minute of a lot of action and the next one five minutes of 'talking heads'. Both parts would have the same average bitrate but the action one would look really bad. If they were included in the same encode process, the bits would be used much better.

Yes, Boulder, you're right. Do you know if there's some plan of making DVD-RB encode the whole movie in the future, instead of cell by cell?

__________________
AudioSlave

I don't know if that's even possible to do..as the rebuilding is done cell-by-cell, it might be impossible to cut the stream and use the parts to rebuild. It might be possible by forcing a keyframe to the frame where the cell begins and then cut accordingly, but it might also be a very unstable solution.

Do you use DVD-RB yourself or do you use another solution?
I 've heard of something called "The Big 3" or something similar... What's that?

__________________
AudioSlave

I use DVD-RB quite seldom, only if I put one movie per disc and want to keep the menus. I always use RB-Opt and one-pass variable encoding. I usually put two or three movies per disc so mostly I do manual work.

Big three is a complicated and advanced process, you can read about it at doom9. I've never tried it myself.