[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Custom Curves

>Since this is done on the primary boards in 16 bits, it shouldn't be a

It's interesting to note that, although 16 bits of data are extrapolated on
the Input board the lookup tables on the Primaries throw away the two least
significant bits for their input.
I don't remember how/if  we get back to a full 16 bits later.  The busses
are 16 bits wide...but do they actually carry valid data of this width?

>Actually there are four lookup tables. Two sets of inside and outside
>tables, ...

Right.  I was only talking about the "live" part of the implementation.

>The Y Matrix board is where the black (16) and white (235) levels are
>clipped, the Y channel has only one DSP and therefore when you adjust the
>black clip or white clip levels you see the DSP being updated (green/purple
>flashes, etc..) on the video output.

Let's not confuse the Custom Curves processing mechanism with Output
Clipping and Formatting.
Output Formatting is where the internal zero-origin numbers are converted to
what is appropriate for the desired output format:  in eight bit 601, black
would be shifted up to 16 from 0, for example.  The other part of formatting
is putting pixel data back in the correct order (Cb-Y-Cr-Y, etc).  This
happens on the Output board, not before.

Output Clipping does happen on the Y Matrix board and takes the form of a
lookup table loaded by the DSP.    I guess it could be a matter of
semantics...the flickering you see is the lookup table being written to
"live".  The fact that there's only one DSP has nothing to do with it.  The
multiple DSP's on the Primaries boards are there because each one takes care
of each channel's (YRGB) inside/outside and live/offline lookup-tables.

>You could also run a paint program on the SGI and take the digital output
>into the daVinci key channel input..

A lot of really neat stuff is going to start happening in Telecine bays when
the sort of integration you are talking about is better understood and
becomes more popular.  I wonder just how much of this "doodling" clients are
going to tolerate.  I think that the Flint/Fire/Flame/Inferno folks have
done a great job getting clients comfortable with the idea of waiting for
stuff they used to be able to do in real time.  Now those boxes are
substantially faster...but the word "render" is now part of almost any
client's vocabulary and not necessarily a bad word.

BTW, I'm not putting down the FFFI packages.  We all know you can't match
their versatility and range of functions.
...but you still need several Onyx's to be able to do what a little-old
8:8:8 can do in real time with a bunch of clever silicon circuitry.

BTW #2.  Since the original post I've looked at the various tricks suggested
to achieve layered gamma curves and, as far as I can tell, it cannot be
done.  Once you have a window or key filling the whole frame the only gamma
curve or correction you are going to see is the "inside".  The only hope for
layering gamma curves is to use the Colorgrade gamma to achieve a first
grading and then do a Custom Curve for anything more complex.

The good news here, as someone pointed out, is that I can now go to Fry's
and spend money!


Thanks to Complete Post L.A. for supporting the TIG in 1998..
No product marketing allowed on the main TIG.  Contact rob at alegria.com
957 subscribers in 36 countries on Thu Mar 19 04:26:49 PST 1998 
complete information on the TIG website http://www.alegria.com/tig3/