Avisynth: Increasing Compression Through Preprocessing - Page 2

SansGrip · #21 01-14-2003, 11:51 PM

Quote:

Originally Posted by kwag

That looks perfectly good for one CD

It's pretty good, but it's no Signs -- I'll keep tinkering with it

.

kwag · #22 01-15-2003, 12:01 AM

Well it seems I'll be re-encoding "Monte Cristo" with a CQ > 70 at 528x480

I just ran a sample at 70 and was ~1MB lower then predicted. And I also had 1 blocks overscan, and not 2 as I usually do for 528x480. So I figure the CQ is going to be ~74-75, and that's a hell of good quality for 528x480

-kwag

SansGrip · #23 01-15-2003, 12:07 AM

Quote:

Originally Posted by kwag

the CQ is going to be ~74-75, and that's a hell of good quality for 528x480

It sure is. What's the movie length/AR? What script?

kwag · #24 01-15-2003, 12:11 AM

Quote:

Originally Posted by SansGrip

Quote:

Originally Posted by kwag

the CQ is going to be ~74-75, and that's a hell of good quality for 528x480

It sure is. What's the movie length/AR? What script?

131 minutes, 32 seconds. Here's the script I'm using:

Code:

LoadPlugin("C:\encoding\MPEG2DEC.dll")
LoadPlugin("C:\encoding\fluxsmooth.dll")
LoadPlugin("C:\encoding\GripFit_preview.dll")
LoadPlugin("C:\encoding\blockbuster.dll")
LoadPlugin("C:\encoding\legalclip.dll")
LoadPlugin("C:\encoding\sampler.dll")
LoadPlugin("C:\encoding\vobsub.dll")
LoadPlugin("C:\encoding\convolution3d.dll")
LoadPlugin("C:\encoding\dustv3.dll")

Mpeg2Source("K:\COUNT_OF_MONTE_CRISTO\VIDEO_TS\count.d2v")
LegalClip()
vobsub("K:\COUNT_OF_MONTE_CRISTO\VIDEO_TS\VTS_01_0")
GripFit( width=528, height=480, overscan=2 )
FaeryDust() 
Convolution3D(preset="movieLQ") 
GripBorders() 
LegalClip()

Sampler(length=24)
## MPEG size = ((Total frames/MovieTimeInMinutes)/24) * MPEG sample file size ##

Looks like final CQ is going to be 71.5, as 72 was just barely over.

Edit: Aspect wide screen 16:9

-kwag

SansGrip · #25 01-15-2003, 12:16 AM

Quote:

Originally Posted by kwag

131 minutes, 32 seconds. [...] Aspect wide screen 16:9

That's a very high CQ for a >2hr 16:9 movie... Mind you, it should be fairly compressible because it's quite dark and mostly low-action. Gonna post a sample?

kwag · #26 01-15-2003, 12:19 AM

Quote:

Originally Posted by SansGrip

Quote:

Originally Posted by kwag

131 minutes, 32 seconds. [...] Aspect wide screen 16:9

That's a very high CQ for a >2hr 16:9 movie... Mind you, it should be fairly compressible because it's quite dark and mostly low-action. Gonna post a sample?

You bet

. 5 minutes to end the sample ( FaeryDust is really slow

)
Then I'll cut a sample and post it here. Don't go to sleep yet

-kwag

SansGrip · #27 01-15-2003, 12:25 AM

Quote:

Originally Posted by kwag

FaeryDust is really slow

Very. However, Steady is already working on a new version (mostly rewritten, from what I can gather) that uses a different algorithm for motion estimation. Should be much faster.

Quote:

Don't go to sleep yet

I'm going to have to watch that sample tomorrow, because I'm dead in my chair here. Have a good one -- see you in the morning

.

kwag · #28 01-15-2003, 12:31 AM

So it will be CQ=71

Here you go: http://www.kvcd.net/count-sample-528x480-cq71.mpg
That's for one CD.

-kwag

GFR · #29 01-15-2003, 04:22 AM

Quote:

Originally Posted by SansGrip

Quote:

Originally Posted by kwag

(haven't seen the movie yet, just the sample

)

It's not great

-- nowhere near as much swashbuckling as there should be

.

Agreed - not great.

But, DO read the book! It's awesome!

SansGrip · #30 01-15-2003, 07:44 AM

Quote:

Originally Posted by kwag

That's for one CD.

Very nice... Man, I'd forgotten how dark that movie is

.

SansGrip · #31 01-15-2003, 12:09 PM

If you do a StackVertical compare of a sample done with, say, Flux against a sample done with FaeryDust and C3D movieLQ you'll notice quite a significant loss of detail in the latter. That's where all the compression comes from

.

I'm thinking that movieLQ is a good choice if you really want to get something on one disc and don't have an HDTV, but otherwise movieHQ might be more suitable.

In fact what I'm really thinking is that we need a filter to do an edge detection then smooth non-edge areas and sharpen the edges simultaneously. That would be nice. But what to call it?

SansGrip · #32 01-15-2003, 12:13 PM

Look in particular at Gibson's face and jacket:

The top was done with Flux and TemporalSoften. The bottom with FaeryDust and C3D movieLQ.

kwag · #33 01-15-2003, 12:17 PM

What about Flux + FaeryDust
Or TemporalSmoother + C3D
Or FaeryDust + TemporalSmoother

(

)

-kwag

SansGrip · #34 01-15-2003, 12:44 PM

Quote:

Originally Posted by kwag

What about Flux + FaeryDust

Well, they're both denoisers, so you don't really need to use them together. Faery should catch all the noise that Flux does, and all the noise it doesn't, too

.

Quote:

Or TemporalSmoother + C3D

And these are both smoothers rather than denoisers. I think we want:

Flux or FaeryDust

and a smoother, be it TemporalSoften, TemporalSmoother, C3D, or my as-yet-unnamed smoother/sharpener combo filter

.