On Fri, 18 Mar 2005, E.Chalaron wrote:

> Hello Steven, thanks for the mail

        You're welcome.

> Well I tend to do the following to get the reels under Kino :
> 
> (find . -name \*.pnm | xargs cat) | pnmtoy4m -F 25:1 | y4mdenoise  -I 0 | \ 
> y4mscaler -v 0 -I active=YYYxYYY+AA+BB -S option=cubic -O infer=exact -O 
> size=720x576 -O chromass=444 | y4munsharp | y4mtoppm | ppm2raw -v 0 -a -2\ > 
> ../reelXX.dv

        Ok.  I wonder - is ppm2raw creating 4:1:1 or 420paldv?  

> I found out that using the Edge Sense Bayer filter brings some anomalies as a 
> kind of tartan pattern in (too) bright spots. The sinc:5 was ringing over it, 
> so went for -S cubic in y4mscaler

        That's why there's a choice of scaling kernels - pick the right one
        for the task.

> >     ...terrible but  I do think all that resampling/conversion will exact
> >     a price on the quality.
> 
> Sorry I don't get it... Don't forget that I denoise 1300*1100 ~ish frames 
> before bringing them down to DV size.

        The thought I had was that going from 4:4:4 to 4:1:1 to 4:2:0 involves
        2 resamplings of the chroma and resampling never improves quality.
        
        More importantly is the lossy conversion to DV.  Think of it this way,
        a 720x576 4:2:0 frame is 720*576*3/2 bytes or 622080 bytes but a PAL-DV
        frame is 144000 bytes.  There's a "jpeg-like" (lossy) compression 
        taking place.
        
> >     Any particular reason you're converting to DV?  That's a lossy process.
> 
> Well, I have to do 2 things, swap images Lef to Right since I look directly 
> into the projector gate, then I need to add extra frames for 25 fps. Most of 

        Ah, ok. I thought perhaps you were doing the left to right swap as
        the data was being captured or shortly after that.  At one time you
        were thinking about using yuvfps to do the rate conversion or perhaps
        writing a custom pulldown program.

> >     No -q and no --cbr means that -q is defaulted to 8 - and that can be
> 
> So you reckon I can boost it.. -q 6 is still blocky. Is it a linear scale ?

        Yes, I think you can boost it.  Try 4 and see if that works a little
        better - but what's really needed is some filtering that I don't know
        how to do with the mjpegtools filters.  

        It's linear up to 8 and diverges after that.  The 'q' range of 1 to 31
        maps to 1 - 127.  Values over 8 are rather poor quality and not used
        much (I know I've never used them ;)).

> Well I'd like to go over 7500 but I'd rather stick to something that a 
> dvdstandalone player will not spit out.. 

        The rate control seems to have improved in the cvs version of mpeg2enc
        so I think you'd be fairly safe with 8000.  Not sure if it'd make much
        difference.

> >     Another thing to try is using "--cbr" instead of "-q" - that will
> 
> I thought --cbr was evil

        I don't care for it but that doesn't make it evil ;)

        It gives more predictable filesizes by allowing the encoder to 
        dynamically vary the effective -q.

> It mostly happens on dark ~ish scenes.... I tried to add a bit of gain, it is 

        I think if you look at the ppm files with something like the GIMP
        or Photoshop you'll see that what looks 'black' to the eye is really
        varying  levels of dark color which gets magnified when going thru
        the quantization/encoding process.

        Two things I've found that really help a LOT with the blockiness in
        dark scenese are 1) look at the data with a waveform monitor and
        check the black levels - if they are consistently higher than 16 then
        'crush the blacks' slightly and 2) "desaturate the lows" - the human
        eye does not see color well in the dark but the encoder does and the
        blocks you see are a result of that.  Removing some of the color info
        from darker areas works wonders.  For these types of things I use
        Final Cut Pro.

> Will try the options of --cbr or -q 4 without -H
> I'll post some feed back on that...

        Ok - have fun and let us know how it works.

        Cheers,
        Steven Schultz



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click
_______________________________________________
Mjpeg-users mailing list
Mjpeg-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mjpeg-users

Reply via email to