Hi jorel,

I think I know what you mean. It's the extremely high chroma values that are applied on very low movement scenes. I'm fixing that right now :)

-kwag

thanks for atention Kwag!
:D

dredj a few minutes in his second post, see the same :
http://www.kvcd.net/forum/viewtopic.php?t=4729
:!:

Thanks for all of you with eagle eyes out there :)
I was able to see the effect CLEARLY on a specific scene on K-Pax. Here's the fix:

The previous values were just plain too strong :!:
There shouldn't be any "Bleeding" of chroma now. Period :!:
I'm updating the script page right now.

-kwag

What now, are we in trouble again :?:
good thing i hanen't encode enything yet. :twisted:

Don't worry, we nailed this one in less than 24 hours :lol:

-kwag

yes dear friend, right between the eyes.
:lol:
really better(my opinion)
:D

and when i load the script in vdubmob it run soft,
the system runs easy too and the image is cool without "breath" :!:

the best...the prediction CQ encrease 3 points.
:wink:

@ Osvaldo,
i did a encode when i see it and have to do all again.
i have 9 8O
movies to encode, my 2 hds are full of vobs.(100Gb)
:lol:

Hey Kwag TemporalCleaner sims to do a good job :!:
TemporalCleaner (ythresh=5, cthresh=10)
Im becoming a filter frik now :mrgreen:

That's exactly how I found the problem. It looked really nasty on a scene in Vdub, where the person sits down and the camera stays still. So it's a vertical scroll. So I tunned the values until the effect was gone, and then I lowered it about 10% for insurance ;) Same here :)

-kwag

8)

everything is better!

more CQ, more image quality, the system run soft and faster too.

big goal Kwag :!:
:D

LOL
 

8O 8O 8O 8O 8O LOLROTF 8O 8O 8O 8O 8O
Just finished encoding K-Pax with the script update.
I used ToK and "Fast" prediction, for a CQ of 64.18

Wanted video file size by MovieStacker: 714,303.19KB
Encoded file size: 714,352KB :mrgreen: :mrgreen::mrgreen::mrgreen:

I don't think we need to calculate the % precision this time :lol:

-kwag

Re: LOL
 

Wha :?: :?: :!: :!: Man I hope my results are as good as yours. I'm going to try out the new script right now. Kwag you still using ToK 0.0.5.2?

Re: LOL
 

Yep :!: :)

-kwag

Kwag,
see my results and if i'm using the right adjusts in ToK

CQs predictions from ToK0052 with the new script!

using 1st group of passes:
precidion% 0,50
speed up by 1 time
:arrow: CQ: 59,639 :wink:

and using new faster prediction (Tenra)
:arrow: CQ: 59,638 :wink:

right?
you only check the (Tenra) to got your results :?:
:)

Hi jorel,

Yes, I only check "New Faster Prediction", set Precision to 0.5% and Speed Up by 10.
For full prediction, I uncheck the faster prediction, and also uncheck the 2nd group of passes. Then I change the first group speed Up to 1.
For all, I leave the factor for prediction set to 1.0.

-kwag

thanks... :D

testing all ways now!
:wink:

Well what can I say :!:, here's a sample from the last encode :)
I think this just can't get any better ( at least for the next 48 hours :lol: )
This looks almost like ( or like? ) my DVD :mrgreen:
www.kvcd.net/k-pax-fixed-adaptive.mpg
Comments always welcome :cool:

-kwag ;)

Kwag
i download the k-pax sample and see
good quality colors with great matizes, amazing sharpness too.

right as you wrote
"I think this just can't get any better ( at least for the next 48 hours :lol: )"

:D
for me too.

8O

How is it possible that CQ goes up when you lower the filter values?? 8O

Probably because the values were going up so high, that the color bleed caused chroma/luma shifting and ghosting, and the image was actually dirtier :)
Now that the picture is extremely clean, the CQ goes up :D

-kwag

Re: LOL
 

Well I wasn't as luck. Came up 30MB short on a 800mb CD encode with the new script. Looks great though, so I'll settle for a little loss in space.

Kwag, are you using 1 sample per minute and the auto sample size for your prediction?

Kwag,

Does the latest changes on the optimize script applicable or will work best only for a very clean material like DVD?

Thanks,
Dredj

Re: LOL
 

Was that with "Fast" predition or full prediction :?: Yes I am.

-kwag

All I can show about how things are going is: :ole: keep up the good work ppl shows what teamwork can do.

:fahr: (had to do this one seen it and thought it was funny)

Jorel sent me that one :arrow: :fahr: :D

-kwag

Re: LOL
 

That was with Full prediction set to .5% precision. When I used fast i got a CQ 5 pts lower

Man this stinks. A new movie just finished and was 7% short of target for 1CD. 50 wasted MB bah :evil:

Could it be because these videos I'm encoding are hybrids (about 70% film and 30%) video and I'm using telecide and decimate to bring everything to 23.976 fps? Could something like that throw prediction that far off?

hey audi, you're not the only one getting short video files! Kwag are you the only one getting accurate prediction? Sometimes my movies are only 90 min long so I usually do prediction with 2 samples per minute. In my latest encode:

Required Video Size: 732,032,000
Encoded Video Size: 693,986,212

That's 38 Mb short, or 5.2% off. Frankly I don't know what this means but a prediction factor of 1.0 for me is NOT doing the trick.

@audi2honda,

You bet it can :!:
You'll probably need to sample a wider "window" to get a better accurate result, and even that I'm not sure if it will be accurate enough. All my encodes have been progressive. I'm pretty sure that if your material was 100% telecined, and you use telecide and decimate, you'll still get accurate resulte. But mixed 8O :?: :!: :!: Not sure.

-kwag

Ok I'll try one of my progressive DVDs tonight when I get home.

J-Wo are you having problems on clean progressive material or just other material?

My current project are DVD versions of a television series that have all sorts of hybrid nasty interlaced stuff in them. Telecide and decimate produce wonderful results, but I guess that could be throwing prediction.

Kwag do you recommend more samples per minute or a longer sample size? or both?

You're probably going to have to do 48 or to 60 frames per sample, and then maybe at least 100 samples per movie. And still I'm not sure of the results, because of the randomness of the mixed nature of your source :roll:

-kwag

It's a randome mixture but after frameserving from AVS with the telecide and decimate functions isn't it all clean 23.976 when it gets to TMPGEnc? I would think in that case prediction would be accurate because TMPGEnc doesn't know since it's allready been converted or IVTC'd

Yes but because the film is not going to be smooth ( some parts will be smooth and some will be jumpy ) that will throw off any prediction for sure :!:

-kwag

True True. Thanks kwag I'll try one of my regular dvd's tonight to see what happens.

audi: yes, my sources are 100% film and all progressive. They are from clean DVD or SVCD sources. Kwag, any idea what is throwing off prediction for me? I almost want to revert back to before you got MA with full linearity

I have no idea :!:, but I would like to hear more success/fail stories from many people about this. So far I have encoded three different movies, and the results have been just excelent. I still prefer using the full prediction mode, even if it takes from 20 to 30 extra minutes.

-kwag

I was off my self today by 30mb got a 670mb file when it should have been ~700mb to go with a 95mb audio file, use fast pred. maybe we have to chainge presicion to .10 w action movies :roll: . But one thing i did was cutting credits at the end don't think that would throw off pred. though. :idea:

Hey guys, I was having some trouble with prediction too, my video files were also 30-40megs short or so. I think I've gotten around it by changing the # of frames per sample from 24 to 48, and the # of samples per minute from 1 to 2. I'm also following Kwag's advice and doing full prediction, which always seems to increase CQ by a liiiiiiiitle bit vs. fast prediction. But there it is.

Well, today I encoded the same movie (K19) twice, but after the first encode, I changed the prediction to full without exiting ToK. I did this on purpose to log the "fast" CQ and the "normal" CQ on the same log. Here's the complete log:

Obviously the fast prediction is under target most of the time. And look at the difference on CQ values.
With fast, CQ = 62.584 and file size = 676,918,065
Witn normal, CQ= 62.951 and file size = 696,413,210

Look at the difference in file size, 19,495,145KB but on CQ, it's only 0.367 :!:
Again I got 0.964% ( less than 1% again! ) accuracy with the long prediction.
That's why I'll wait the extra time for the longer prediction :idea:

-kwag

Yes i see a diference of 20mb in file size, but :?: is there a difference
in Quality between 62.58% n 62.95% :?: i dont think so but it would be great to be closer to the target w fast prediction. :)