Re: DNG as an ISO standard

Wolfgang Weisselberg wrote:
Alan Browne <alan.browne@xxxxxxxxxxxxxxxxxxxxx> wrote:
Wolfgang Weisselberg wrote:
Alan Browne <alan.browne@xxxxxxxxxxxxxxxxxxxxx> wrote:

I think you missed the point:

Nope, I got what you said. I just don't believe it's true.

in camera raw -> cr2 can be replaced by in-camera raw-> dng

True.

at little cost in cpu cycles.

Questionable. Better compression almost always cost exponentially
more CPU and RAM for 1/ln shrinking gains, asymptotically reaching
it's entropy.

That's the point.

The real point is you don't know and secondly that if reduced to
optimized hardware rather than cpu cycles there would be an order of
magnitude gain in cycles as well as power. Esp. as in a h/w solution
the process can be (despite your note below) highly parallelized.

And who knows, it's not out of the question that direct to
DNG might take less cycles than raw to CR2 in any case...

I ... doubt that.

The doubt is reasonable in either sense. What cameras do today in camera dwarfs what common PC's could do 15 years ago.

15 years ago (1993) the common PC had a 386 with 1-4 Megabyte
RAM and a 20-60 Megabyte MFM harddisk (with a separate
controller). Internet was completely unknown to Microsoft.
SPAM was not a common phenomenon.

That doesn't mean a single thing, though.

Of course not as cameras have neither a harddisk (unless in the storage)
or online storage.

The point is simply that as time goes by, more can be done in less chip
space at higher speed.


This trend is not about to stop

... but the hardware isn't exactly growin that much these
days, instead we get more and more parallel processors.
Which is good enough for servers or parallelizable tasks.
Compression isn't.

Of course it is. That common algorithms's don't do it does not mean it
can not be done. Read the raw into a chip in (eg) 16 "stripes" of the
raw data (RAM -> chip) and compress each strip of data in parallel. Or
32 stripes or 128 or...

This would be insignificantly less optimal than a serial compression in
terms of storage space, but the processing time would be dramatically
faster.


and specialized algorithms can be reduced to hardware for higher energy and time efficiencies.

True --- but better compression is getting much faster more
CPU-hungry than the hardware gets faster.

More reason to implement massive problems to a dedicated chip.

Essentially, raw -> DNG can probably be done without a significant difference in processing than raw -> cr2 and that that difference is probably worth the additional compression.

If your assumption about the RAW->DNG (with the compression you
observe on your PC) was true, I'd agree wholeheartedly. However,
the compression part in RAW->DNG could also be done as RAW->CR3.

Standardization is the goal. IAC, I can find no references to CR3 as an
existing format.

If DNG gets accepted widely, you should see your RAW->DNG
w/compression in 3-5 hardware generations --- by then the CPU
and RAM power will probably be great enough or the (marketing)
drawbacks too strong otherwise.

Hasselblad, Leica, Samsung and Ricoh simply went with DNG as their raw
format. And while each of these camera makers has its pluses and
minuses I haven't heard any complaints wrt their raw format.

Eg: it's here. The issues is Canon, Nikon, Sony, Pentax... et al
steadfastly holding on to their proprietary formats.

Akin, if you will, to each specifying a different film size.


It would be easier to have the camera store them as DNG.

even if the compression wasn't as good?

Again, the point is standardization which reduces workload and allows
for more cross s/w and platform work.

And again, CR2 is less compressed than DNG and I can find no such thing
as CR3.

--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: Remove FreeLunch.
.

Relevant Pages