will do... thanks guys!

Thanks SansGrip,
Now I know why my predictions have not been accurate the old trying to fit 25 frames into a 24 frame Gop trick.

well... my predictions are off!

with my scale factor at 0.85 and using Sampler(length=24), i am producing a file just a bit larger than expected.

final file size was 820MB... very close, but still no cigar. i might go back to the other way:

###------------------- Start Of File Size Prediction -------------------###
#
IL = Framecount / 100 # interval length in frames.
SL = round(Framerate) # sample length in frames.
SelectRangeEvery(IL,SL)
### Predicted MPEG size = (( Total frames/MovieTimeInMinutes)/24) * MPEG sample file size ###
#
###-----------------------End File Size Prediction----------------------###

any thoughts on this?

I believe the latest scale factor is 1.0, i.e. no adjustment necessary. kwag'll correct me if I'm wrong :).

You are correct :D

-kwag

Is there any need to change the sample points value in KVCDP?

Jeez, I just got back home yesterday from a two-week holiday and you've already making me feel like I've been away for two years!

Yep. It now needs to equal minutes-in-movie. It'll do this by default in the next release.

Ah, you mustn't go on holiday! I assume you've read all about the new GOP, experimental Q matrix, CQ mode for 528x480 and above... :D

...and don't forget to set the error margin to 0%. We don't seem to need it any more :).

Well, if one gets free food and drink for the two weeks, one must do what's right :wink:

I've read most of the new posts, this all seems to be going somewhere again. Luckily(?) , I use the larger resolutions rarely so I haven't had to experiment with CQ yet. I'm currently trying out the BETA-1 notch matrix - stuffing a 2-hour PAL movie on one CD at 352x576. The prediction sample looked almost as good as the source, the original itself is quite soft as it's an old movie.

Well, why didn't you say that in the first place?? ;)

That threw me for a minute. By "larger" you mean 352x???, right? :)

It's this kind of thing that makes me think my tests of the new matrix were somehow flawed. I think I'm going to run some more today :).

352x288 is low, 352x576 is ..um.. 352x576 and anything above that - meaning KVCDx3 and 704x576 - is considered a larger resolution in my small mind. I mostly use 352x576 for TV capture encodes these days.

Which reminds me, they just started a Star Trek (the original series) rerun on a cable channel. It'll be a good test for the templates, 79 episodes of about 50 mins each, maybe 40 CDs?

I just reread your post and noticed the word "rarely". My mistake ;).

Sounds about right. Since it's full-screen I wouldn't think you'd get more than 100 minutes on one disc at 352x576. Though you never know until you try... ;)

well, i used a scale factor of 0.90 and my total 750MB. im so confused because before, my scale factor of 0.85 came out to 820MB.

im freakin out man!!!

It's counterintuitive ;). As you approach 1.0 your total size will drop. I recommend just using 1.0 and forgetting about the mechanics of it ;).

Hi there,
I believed I have seen somthing about adjustment in sampler when using in CQ mode, and I believed someone said the sampler is not that accurate when reach a high level of CQ (say 85). So, is there an update in using sampler?

I did the movie "Signs", I used sampler (length=18) because it was a KDVD, the acutally file size is 1.3104 larger than predicted (predicted=1807MB, actual=2368MB). I used sampler-0.2a, any idea please?

It's not really Sampler that's inaccurate with a high CQ, it's apparently the method itself.

This might be having some impact. I saw similarly inaccurate results with some values when trying to determine the best sample length with the new KVCD GOP.

One thing you could try is using a length of 24 (even though your GOP size is 18 ). I'd be interested to know if this is more accurate.

Because I've never used the KDVD template I'm not really sure what to suggest. You could perhaps switch to CQ_VBR mode and see if that's more accurate, or use 1.3 as your scale modifier with CQ mode.

I'm afraid more testing needs to be done with high CQ levels. On the other hand, think of the glory that will be poured upon you when you figure out a solution :mrgreen:.

Hi there Sansgrip!

I finally managed to register on this board! I think this is the 5:th time or so... I never got the mail sent to me that allowed me to activate the account... But now it - for some reason - finally worked!

I have been enjoying the information on this board for weeks though, and I'm also using some of your filters.

I was messing round with CQ long before you guys figured out that it produced better results (in my opinion). I'm also working on my own prediction system which seems to be different from yours.
For this I've been using your Sampler() filter, but I would like to get a more accurate testresult.
I was thinking about doubling the sample, but it seems I can only do that by doubling the length of the clip. I would prefere to double the number of clips, but I can't seem to get it right... Is there anyway to do this?

Other things I would like to share now that I'm finally "visible".

I removed blockbuster since it made the background blocky (walls and such)

I removed nomosmooth since it took too much time for little result when converting.

I also removed fluxsmooth, but I'm thinking of adding this again if I find that the time to convert isn't doubled. (that's why I removed it in the first place)
Yes, I know I should have checked this out before, but right now I'm struggling to convert both Gangs of New York and The Two Towers, and they have got to be ready by tomorrow.

If I only could get the prediction more accurate I would allready be done with them, but I had to change the factor this morning, and start over... It was too big, not much, but enough to not fit on a CD.

Anyways, thanks for reading, and thanks for your great filters.


/Uffowich

Yep. By default Sampler takes minutes-in-movie samples, so you could simply override that by saying:

Sampler(samples=minutes-in-movie-times-two)

Yes, Blockbuster doesn't seem to work as well with CQ mode.

It shouldn't, because Flux has been MMX-optimized for several versions now. Give it a try.

I think that's what we're all struggling with right now -- finding the right "correction factor" for the file prediction formula. At the moment it looks like 0.98 or 0.97 might give good results.