This article should be read by everybody

Started Dec 12, 2013 | Discussions thread
TrojMacReady
TrojMacReady Veteran Member • Posts: 8,729
Yes, let's enter the fallacies.
1

K E Hoffman wrote:

TrojMacReady wrote:

K E Hoffman wrote:

Its nice to believe .. life is better with Tinkerbell and Santa... and adding complexity to sensors with no noise cost.. But Tinkerbell is a lady on a wire.. Santa has parents to do the heavy lifting and smaller added pixels to a sensor of the same area and generation has a cost.... noise.

There is no reason to believe, there are reliable measurements and tests supporting my point.

But the screenshots of tests you provided (and provide below) have a whole lot of issues. Dpreview does not normalize for ISO differences (meaning different real sensitivities vs indicated sensitivities), nor output size. Nothing reliable to conclude from that about noise at an image scale for a given physical exposure.

The table you posted below is from the more than inconsistent PopPhoto data, based on pixel leveloutput rather than equally sized output and the kicker: OOC jpegs. Even more problems than mentioned above. They rated the RX100 as a noisy camera, even compared to much smaller sensor cameras.

Then smart people swat the noise..each new design so we get more MP and still get images we like mostly.

You can see some of this cost in the 60D to 70D The sensors or almost the same image resolution. And the 70D is one generation better.. But it takes a hit on noise at higher ISO very quickly and with a big jump because IMHO the gain is picking up the noise from the increased complexity of the 40MP sensor sites. So it almost hits "unacceptable a full ISO stop sooner for a "20MP" sensor At lower ISO the dual pixel might even be helping it filter out noise . See Below... The 40 MP of actual sensors adds complexity and noise... and that is noise cost that is outside the theoretical "surface area all has the same noise argument". Which has to ignore the complexity noise cost. of added transistors circuit paths..

Proven wrong by my example posted in the previous reply that you cut out and ignored (why?). Same generation cameras: 700D vs 70D, one has more than double the amount of subpixels, same amount of noise at extreme ISO's, one might even argue there's less in the 70D shot:

However I do agree if you scale your 24 MP images.. to 8x10.. Tinkerbell has no wires and you might even miss Dad standing behind Santa.. just don't enlarge or crop those images because reality will hit you at some point.

Why enlarge one and not the other? Why change 2 variables again? If you enlarge the lower resolution shot equally, noise will be just as obtrusive, if not more visually obtrusive as the "blobs" will be larger, which in practise usually jumps out more.

60D note where the noise is at ISO 800

70D note where the noise is at ISO 800.. 18MP complexity to 40 MP complexity jump even on a new generation sensor

See above. Non normalized output, OOC jpegs (completely different jpeg engines and different NR approach). Non repeatable test.

Here's a Sony sensor comparison for you. 20 MP (RX100II) vs 24MP without OSPDAF (D610). The latter is 7.4 times larger, so let's virtually "stitch together" 7.4 RX100II sensors to get a 148 MP sensor. Or crop similarly sized (sensor wise, not image wise...) areas from the sensors. Comparing them again at a normalized output (for example 24MP) would mean resizing the 148 MP file by a factor 2.5 per axis.

To simulate the above scenario or crop similarly sized areas from both existing sensors, we can take grey patches of equal brightness from existing RX100II and D610 files and compare them, based on the above normalization and resizing process.

Middle grey, "indoor" Dpreview scene, center of the frame, just above Dpreview logo:

D610 ISO 6400

RX100II ISO 6400

D610 ISO 3200

Shadows, "indoor" Dpreview scene, bottom of the frame, left to the bottles:

D610 ISO 6400

RX100II ISO 6400

D610 ISO 3200

I added the ISO 3200 from the D610 because that's where real physical exposures seem to match up, yet the noise is still quite similar per unit sensor area, compared to the RX100II.

Conversions using Adobe, no NR or sharpening, resampling done with Lanczos. A test that is repeatable, the files are there.

Conclusion based on actual image test: existing Sony sensors with pixel densities over 6 times higher than the current best 24 MP FF sensor, perform similar per unit sensor area in terms of noise, even in shadows and low light up to at least and indicated ISO 6400. Which means that Sony is technologically already capable of building a competitive (low light, at least up to ISO 6400) 148 MP FF sensor. And with the above (520% increase in pixel density) in mind, it makes the whole discussion about mere 50% or even 100% increase of current MP counts, a bit of a waste of time.

And as to questions "why don't they?". Processing power, buffer, storage and shot to shot times (which all adds $$$$$ once you start countering) being the most obvious reasons.

So.. if we always scale like you do.. and like to shoot gray cards and closets they are the same.
Got it.

You can scale to any size you want or not scale at all and print at the same size. It's called normalization, standard process in any decent comparison.

As for the rest of your comment, I'm not sure how that fallacy contributes to this discussion. It's a test that mimics low light conditions (lots of shadows, typical harsh indoor light source lacking in the blue channel which results in more visible (blue) noise). Nothing more, nothing less.

The DXO system is interesting we pay for 24 or 36 MP... the scale it to 8MP. Problem with that is..

8MP is not even close to hitting basic 35MM film resolution. Again this is a great standard if you bought your camera for web pages and 8x10 prints.

So yes if you scale a modern camera to a substandard resolution they are all the same.

I'm not sure how many times this has to be repeated (and will be ignored again), but they could have picked any resolution, the results would have been the same. It just ended up being 8MP. Same goes for you, pick any resolution or print size as long as it's equal, for not entering another variable.

I love my Sony.. but this is a silly rationalization to try to pretend that an A77 and D600 are the same.

Nowhere did I or anyone else in this thread even suggest anything close to that. Maybe you're mixing up posts from other threads or forums?

I've explained how efficiency per unit sensor area and sensor area is what it's really about and the lack of correlation between the former and pixeldensities in practise. Nowhere did the A77 enter the equation.

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