An extremely low compressibility check result with "Pi"

[curna]

Hi, I'm trying to encode "Pi" (It's a black and white film).

This film is only 80 minuts long and I tried to fit it in 1 CD. This let me to an average bitrate for image of 1039 kBits/s. Also, it's a 4:3 film but have some big black bars at the top and at the bottom (The Aspect ratio is 1.713)

When encoding I was using B-frames en GMC.

Here is the script (using Avisynth 2.50) Gordinat-Knot used to make the compressibility test:


LoadPlugin("C:\PROGRA~1\GORDIA~1\mpeg2dec3.dll")


mpeg2source("E:\DVD RIP\PI_SCN\VIDEO_TS\pi.d2v")
crop(2,56,714,456)
BilinearResize(576,336)
SelectRangeEvery(350,14)

With these settings, the compressibility test gives me 1.169 Bits/(pixel*frame).

This means that, using a "normal" resolution of 544*320, the compressibility test is only of 20.4% using 0.239 Bits(pixel*frame) !!!!

Is this really possible ??? It's this film so uncompressible ???

If that is the case, is there any way to modify my script to make it a little bit more compressible or any workaround ???

Looking at the image it seems to me that have a lot of noise, but the film is from 1998. So, I think that this is what the director wanted. Can this have something to do with this low results ??

I've attached a screenshot of the film, so that you can see how it looks like.

[manono]

Hi-

Pi: Faith In Chaos is among the least compressible films ever made. He may have wanted all the noise, but I think it also has to do with Aranofsky having a very low budget and not having enough money to buy decent film stock (it was filmed in 16 mm on a $60,000 budget). In addition, the handheld, always moving camera makes it hell to compress.

So, you might try some of the Temporal and Spatial Smoothers available. TemporalCleaner, TemporalSoften, FluxSmooth and/or Convolution3D will help its compressibility (all either built in (TemporalSoften), or available as Plugins For AviSynth 2.50 (http://www.avisynth.org/index.php?page=Section+3%3A+Filters+and+colorspaces#q3.4)).

There is one good thing about a grainy film and that is that you won't need as high a Compress Test result as is usual for it to still look good. But you might consider doing it for 2 CDs.

[BluDChyLD]

good taste, Pi's a great movie :D You really should try two cds instead. I encoded it at 512 width without any noise filters, and retained the original 2.0 ac3. Try it this way, should give nice results :)

[Dagon]

I've done a 1 cd version of Pi and a 2-cd version. Both used convolution3d at vhs quality preset, I think.

The one Cd version was not bad at all considering.

Sorry I don't remember resolution details but, my advice is that this movie will still look fine even with low compression %.

[aliens3]

Hi !
I have a similar problem.
I have this Mike Oldfield concert, it's only 56 minutes long, PAL interlaced 4:3, but at 1500 kbps, the compressibility test gives only about 40%. I enabled B frames, pshychovisual enh to normal, bilinear resize. I really don't know what to do to increase the compressibility.

Any ideas ??

[manono]

Lower the resolution?

[aliens3]

Resolution it's already 512x384.
I don't know if it's ok to go lower ..

[manono]

Yeah-512x384 is about as low as I'd go also. Concert films are tough, especially if they're full screen 1.33:1. They frequently use handheld cameras, so the picture bounces around and is hard to compress, and they can be real noisy. Perhaps a temporal smoother such as TemporalCleaner or TemporalSoften might help. Or perhaps a combo smoother such as FluxSmooth or C3D might help also.

[N_F]

As long as it's 4:3 I have no problem going down to 448x336, did it recently with Friends season 7. It's still bigger than 576x240.

640x272 = 174080
448x336 = 150528
576x240 = 138240

@aliens3

On some very special occasion the comp. test can't really be trusted IMHO. For instance with very noisy movies like Pi that's mentioned in the beginning of this thread. Same thing may applay to concerts (never done any, so I wouldn't know).

If I was in your shoes I'd probably use C3D (perhaps in combination with temporalsoften) and make a short test clip to see how it really looks.

[aliens3]

Thanks for your help..

I am going to try with c3d and flux.

[tormento]

It's like resuming dead corpses but I've found the BD of this film.

I've tried with SMDegrain but the noise is far beyond its capabilities.

Has somebody tried to get good results?

Here (http://www.mediafire.com/?1f6uc613r8gsvfv)is a small sample.

Didee created a wonderful script here (http://forum.doom9.org/showthread.php?p=1512302#post1512302)and I'd like to get a version with latest version of degrain library.