Some info on the 24 MP Nikon - from a Sony engineer

Started Oct 29, 2008 | Discussions thread
bobn2
bobn2 Forum Pro • Posts: 69,300
Re: expensive strategy

FrankG wrote:

So you are basically supporting my point which is that for all
intents and purposes at base ISO (at least) "High Density" cameras
such as the new Sony A900 can easily stand up to 100% scrutiny.

That statement is broadly right. The issue is, what is the point of 100% scrutiny.

No-one said anything about downsizing the output. We invariably
upsize it, since no-one's keen looking at images 24x36 mm.

?
Who said anything about viewing or printing at 24x36mm?
When referring to "downsizing" I am of course talking with reference
to comparison to 100% view. Effectively if you print both 12 and
24Mipix image at, say, 19x13 you have effectively downsized the
24Mpix image with reference to 100% significantly more than the
12Mpix image.

The use of the phrase 'of course' assumes that the pixel centric view is natural. I submit that it's not. For all practical purposes (apart from index prints) we upsize the image from the sensor (using the word 'size' literally). We may also upsample or downsample, depending on the output medium. Unless one wants to insist on a rule that there will be pixel for pixel correspondence between input and output (which is almost undoable in the real world, anyway), 100% has little relevance to anything, so why should it be used as a reference for 'upsizing' or 'downsizing'?

The issue is, what do you lose by increasing pixel density? The answer is
nothing (except file size and processor capacity).

Purchase cost!

Nothing intrinsic about high pixel density that increases purchase cost (apart from the already mentioned memory and processor requirements). the manufacturers may decide that it is a desirable enough feature to charge a price premium.

That is true for very Low ISO range pictures. In fact we ought to be
able to do prints of double the size from 24Mpix - but it is a case
of how big do we need to go when we can already get large prints of
extremely high quality from 12Mpix?
...

Why 'ought' we. The extra pixel density brings demonstrably extra detail (in terms of more contrast at high spatial frequencies) even at A3 size.

However as for the A900 being only "one third of a stop" worse noise
than the D700 "at equal output sizes" - would you mind pointing to
your evidence for that?

First look at Emil Martinec's tables
( http://theory.uchicago.edu/~ejm/pix/20d/posts/tests/D300_40D_tests/ )

  • the D300 has to stand in for the A900, and adjust for area. This

might not be completely accurate, but since sensor efficiency seems
to go across sensor families (look at the Canon ones) it's a fair
assumption. Now look at Clarkvision

( http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/index.html ) again letting the D300 stand in for the A900. On the sensor efficiency side (from Emil), we find that the D3(or D700) has about 1/3 stop advantage in area efficiency. Since the sensors are the same size, that translates to overall efficiency. Now look at the read noise figures (from Roger) and we find an advantage (more than a stop) to the A900 at low ISO's, with them being about even at high ISO's. This suggests that if anything the A900 will have a slight advantage at low ISO's. this actually triangulates with real test results, where the A900 seems to have a measurable DR advantage over the D3/D700.

I quote:
"Measured noise as you move up the ISO range is broadly the same for
all cameras, though one glance at the crops above should be enough to
tell you that Sony is making increasingly desperate attempts to
control noise through pretty brutal noise reduction as you head up
the scale. By ISO 3200 the result is a blurry mess with little fine
detail - with the added insult of visible chroma noise in the shadow
areas. I think it's fair to say that ISO 3200 and 6400 are firmly in
the 'emergency use only' bracket (of course with 24MP to play with
you shouldn't have many problems at small print sizes).

That's a long way from your third of a stop estimate.

It's also evaluated at a per pixel level in JPEG. I referred you to the thread where a comparison has been done in raw at equal output size. That, as I said, looks to within 1/3 of a stop. You need to be aware that DPReview has a definite party line here, and the text of their reviews consistently reflects this. Many other reviewers are using similar methodologies, and failing to compare at equal output sizes. The consequence of this is that manufacturers are trying to square the circle, producing the extra pixel density that the market sensibly wants, and trying to avoid being slammed in 100% jpeg noise tests by going for absurd amounts of NR by default. Sony does this, and now Canon is too. Only Pentax seems brave enough to go for the pixel density without heavy NR, but their products are getting criticised in reviews and on these forums as a result.

The other thing we don't know is how the read noise of the Sony fares above 1600. If it's poor that could produce poir noise at high ISO's but that woulnb't be a result of high pixel density or a poor sensel design, but poor read chain design - however, the D300 seems to do OK on the same read chain.

Show me an equal output size raw test that backs up your contention that the noise at high ISO's is much worse, and I will concede, but certainly not on the basis of per pixel noise measurements and 100% crops.

-- hide signature --

Bob

Post (hide subjects) Posted by
(unknown member)
mjt
Mel
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow