Saving Private Ryan

[BluDChyLD]

This film is tought to compress! According to gknot's compression test I only get 60% 512x288 with 3 cds! Im using a neutral bicubic filter, b-frames, gmc and PE set on strong.

Any tip on how I can make it compress more?
Thanks
BluDChyLD

[Acaila]

Not really, just smoothing. I believe I heard that movie is made extra noisy to make it more realistic as a true story. So you'll want to get rid of that extra noise as much as possible.

TemporalSoften(1,5,0) for light smoothing
TemporalSmoother(1) for moderate smoothing
Convolution3D(0,4,4,8,8,3,0) for stronger smoothing

Just some things you can try, play with settings if you don't like the results.

[jggimi]

Noise reduction filters!

This though, is the place where transcoding becomes an art. There is a lot of trial and error, and a lot of content-dependant results.

Too much noise reduction will ruin the the images, not enough will only slow down your encoding without giving you much change in compression.

Take a peek at any Gknot-created .avs file, and you'll see in the comments the AviSynth filters and parameters that theWEF chose for "little" "medium" and "heavy" noise filtering.

There are many, many other options available to you. Check out the AviSynth forum, and you'll see lots of additional filtration tools, lots of arguments back and forth....

You can start with theWEFs offerings, and see if any meet your needs.

[BluDChyLD]

using a low noise filter the compression rises by 7%, I'll experiment a little with different settings and see what happens.

[OvERaCiD23]

one thing, turn off GMC. you'll get jerky playback in panning scenes.
as for filtering, try using DivX5's internal pre-processing. it worked quite well for me on some older films (Hunt for Red October, other stuff from late 80s - early 90s). this method involves very little tweaking and nice results (if you don't want to mess with Convolution3D settings). using the default GKnot settings for Temporal Soften usually smear my images, so i'd watch out for that as well.

[neo_sapien]

Here is my old SPR rip info (http://forum.doom9.org/showthread.php?s=&threadid=16169&highlight=My+SPR)

Basically, compressibility tests are deceptive for SPR. 5% just isn't enough info, I gather from my own experience with that movie. I was able to get 68.9%@ 480x272 neutral using SBC @1292kbits/sec, surely you'll be able to do better, what with B-Frames and Qpel. Do a full first pass @ 544x304 w/simple resize, load the full first pass, and see how compressible the whole movie is.

Good luck.

[BluDChyLD]

thanks for the info, i thought maybe the compression test wasn't long enough... But if it was that inaccurate I would of expected a saturated file (the file size was spot on). It's happened before in some films, but usually the compression test is accurate unless the film content changes dramatically (eg light scenes then lots of dark).

I might re-encode it later using a low noise filter and a bigger comp test, but for now im happy (watching through tv out it looks good).
BluDChyLD

P.S Is there anyway to use a DivX 5 first pass for compression test or do I have to one seperate?

[neo_sapien]

In the Load -> Compressibility Test section of Gordian Knot, just press load and select your firstpass.log. It'll work the same way as with selecting a comp test log.

[BluDChyLD]

I think that method only works with DivX3 ;) SBC encodes, I just tried loading a first pass of one my films and it doesn't accept it.

[jggimi]

It will work fine with DivX5, at least, it has the 10-20 times I've done it.

If you rename the log file to "<name>_AAAxBBB.log" (where AAAxBBB was the resolution) then Gknot will know the resolution used for the test. If not, you will get a warning message stating that there is no way for Gknot to know the resolution. You must have the resolution already set to match the first pass, in that case.

[BluDChyLD]

Originally posted by jggimi
It will work fine with DivX5, at least, it has the 10-20 times I've done it.

If you rename the log file to "<name>_AAAxBBB.log" (where AAAxBBB was the resolution) then Gknot will know the resolution used for the test. If not, you will get a warning message stating that there is no way for Gknot to know the resolution. You must have the resolution already set to match the first pass, in that case.

I tried this and although it doesn't change the percent in the compression box it gives me an average quantizer value which is helpful.

The message says -

Average Quant - 3.088: Compressibility Check failed

Is this correct or should I be getting more info?
Cheers

[jggimi]

I don't know, procedurally, why you and I get different results.

I do this today whenever I run a non-standard .avs through a compressibility test, and I used to to this all the time with DivX 5 -- as we all did, prior to having a DivX 5 compliant version of Gknot. Doom9 even had a guide outlining the procedure.

The .d2v project is open when I select the LOAD button. Perhaps that's the difference.

[neo_sapien]

Well if you're getting an average quant of 3 on a 3CD firstpass then that's a pretty high comp test, like 80% or so.

[BluDChyLD]

i'm going to re-encode it again with 3 cds, using low noise filter and a res of 544x304. I'm also compressing ac3 track to a 160kb mp3 file. I'll post my results later and see if I can load the first pass log this time.

[BluDChyLD]

The quality definetly looks better to my eyes, without the noise it's compressed a lot better. The average quantizer now is 3.652, I still have the same problem load the first pass file though. I have it named as "ryan_544x304.log", I don't see what the problem could be...