[opendtv] Re: RGB mania
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Wed, 19 Jan 2005 10:51:44 -0500
At 10:25 AM +0000 1/19/05, Alan Roberts wrote:
>In an ideal world:
>
>1: acquire linearly from the sensor with at least 16 bits, or
>logarithmically with at least 12 bits.
>2: record it without process and never compress it.
>
>Now all we need is massive storage and extremely wide data pipes.
>
>Strangely enough, the digital cinema world is heading this way already.
>
They have already arrived. Now the rest of the video world is taking notice.
Much of the last Star Wars movie was shot and recorded direct to disc
using the widest gamut they could get out of the Sony HDTV cameras.
All of the next Star wars movie will be recorded in this manner, but
to Sony's latest 1/2" dual tape format, optimized to record full
bandwidth signals from the camera to tape (using of all things,
MPEG-4 part 2 video compression).
As for storage and data pipes, you can hook up a few G5 Macs to an
Xsan 64 bit cluster with up to 16 TB of attached storage and run
Final Cut Pro HD today. It's only 8 or 10 bit 4:2:2 YUV at the moment
for multiple uncompressed HD streams (2-3 depending on the
configuration) or up to 7 uncompressed SD streams. And there is a 32
bit mode for HD rendering to preserve all of the nuances that might
be lost to rounding errors.
Unlike Alan, I do not have problems with compression of the wide band
signals, as long as it is non-destructive. Lossless compression in
the range of 3:1 to 5:1 is quite feasible, with significant positive
benefits in terms of storage requirements. This also helps with the
pipes that inter-connect the storage and CPUs. The only buses that
need to deal with uncompressed wideband signals can be inside the
machine doing the processing.
I would add a few things to Alan's list:
1. Sensors that are matched to typical lens "taking" characteristics.
The sensor should be closer to square with say 2K x 2k resolution. It
should only be constrained by the maximum pixel clock that can be
used to get samples from the chip to the processing circuitry. The
operator should be able to define any subset of samples for capture,
as well as the frame rate of the sampling. Thus one could capture
images at variable raster sizes, aspect ratios and frame rates, up to
the maximum pixel clock for the system. This could include real-time
downsamping as well, to take full advantage of the available samples.
E.G. 2048 x 1152 sensor samples could be resampled to 1280 x 720 to
reduce storage requirements, while maintaining the wideband signals
for downstream processing.
2. The ability to set up graticules in the camera viewfinder to
define safe action and overshoot areas relative to the sampling
parameters outlined above.
Anyone other requests?
Regards
Craig
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at
FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
Other related posts:
