RX100 - Sony, why 20 mp?

tomtom50 said:

wb2trf said:

I am generally skeptical about catholicity in dpr posting, about the desire to believe certain beliefs as a matter of creed. The "higher pixel count is forced by marketeers but worsens IQ" story is just such an invented belief and, when wielded with epithets attached as alexzn does, appears aimed at rallying the faithful. I'm not in that church.

Click to expand...

"higher pixel count is forced by marketeers but worsens IQ" is not an invented belief. Many reviewers lamented falling image quality as manufacturers increased the pixel density in point and shoots, and at times manufacturers backtracked.

Click to expand...

tomtom50 said:

Three cases:

Fuji F series

Fuji produced the F10, then the F20/F30, and then the F31fd, all with 6MP sensors. The cameras were heralded as having amazingly good low light performance, far above what was thought possible in a point and shoot. Check the reviews in DPR and elsewhere, especially for the F30 and F31fd. Also check "Fujifilm FinePix F-series" in Wikipedia. Fuji then upped pixel density in the 8MP F40fd, the 12MP F50fd. Low light performance fell with each increase in pixel count, and the F31fd became a cult camera, one of only a few digital cameras that sold for more used than new.

Click to expand...

The assumption here (since we're talking about pixel densities), is that this was a result of choices in sensor design. Based on the RAW files I have seen, it has much more to do with the jpeg engines instead. The RAW files from the F31fd were fine, but not above what you would expect from the sensor size in terms of noise.

Thus this example tells us exactly nothing about the above theory linked to sensor design choices.

tomtom50 said:

Canon G series

The 14.7MP G10 was followed by the 10MP G11. Reviews acknowledged that the G10 could capture more detail in good light, but they generally preferred the better low light performance of the G11, considering it a more balanced camera.

From DPR "Interestingly, at ISO 1600 the G11 appears to be capturing slightly more detail than the more G10, despite having fewer pixels. The G11's saturation levels are better maintained and fine text that is legible on the G11 simple isn't in the G10 shot. Overall these are some of the best high ISO results we've seen from a compact camera, and a clear stop or so better than the G10."

Click to expand...

Again, the assumption here is that marketing decided the jump in sensor resolution from one generation to the next. Fact is, the G10 is the only Canon G series that does not carry a Sony sensor. That alone could explain a good differences. From the G9 to the G11 the resolution did go up (even if only a bit). The G11 was also using a new type of "slow read out" process for CCD chips to improve SNR, as described in the technical paper from that sensor.

tomtom50 said:

Nikon FF

Nikon branched their professional FF cameras. Lower MP models that broke the mold with their high ISO performance, and high MP models that had more detail in good light. Were the engineers deluded when they brought out the D3s?

Were these reviewers inventing a belief? Were the engineers acting on a tradeoff that did not exist?

It is possible that lower read noise and other improvements have lessened the tradeoffs, but that does not mean they have gone away. They certainly existed in the past if we assume engineers at Sony (G11/G12 sensor), Nikon, and Canon knew what they were doing.

Click to expand...

The G series had CCD sensors, different technology to begin with. And without quantifying the influence of differences in read noise, the visibility of that theoretical difference in existing cameras , remains an assumption.

Also, you have to keep in mind that those top end cameras were (and are) all about speed aswell and read out and processing speeds were even more of a limiting factor than they still are today. So the fact that resolution was kept relatively low, could have been determined by that factor alone, while efficiency and read noise were determined atleast in part by the budget for such a high end camera. No one knows.

In other words, it's easy to assume, but without knowing the precise design settings and constraints, it's still guesswork rather than seeing proof for a theory.

Then why did Canon go from 14.7MP in the G10 to 10MP in the G11?

Why did Nikon make the huge pixel D3s and break all high ISO records?

Why did Fuji F series high ISO worsen as they increased pixels?

It seems some are so bound up in theory that they are not impartial. Let's face it, none of us are sensor engineers, so we cannot separate theory for engineering advancements.

Why deny the existence of a tradeoff Canon and Nikon engineers clearly saw?

Sensors are in general getting better as pixel counts generally rise.

I have actually talked with the designer of a cmos sensor used in high speed video cameras (Vision Research Phantom), and he emphatically told me that bigger pixels help DR and sensitivity, ALL ELSE HELD EQUAL. I suspect he knows more than any of us.

TrojMacReady said:

The RAW files from the F31fd were fine, but not above what you would expect from the sensor size in terms of noise.

Click to expand...

I could say look at the reviews, which strongly contradict you. Instead I will note that the Fuji f31fd did not produce RAW files.

Ahem.

tomtom50 said:

Then why did Canon go from 14.7MP in the G10 to 10MP in the G11?

Click to expand...

First because they bought sensors from other makers and the G10 was the only G series without a Sony sensor, see my post above.

tomtom50 said:

Why did Nikon make the huge pixel D3s and break all high ISO records?

Click to expand...

All answered above and I'll add one thing, the D3S was the first consumer camera with a sensor to us AR coatings improving QE.

tomtom50 said:

Why did Fuji F series high ISO worsen as they increased pixels?

Click to expand...

Mostly because their jpeg engine didn't keep up.

tomtom50 said:

It seems some are so bound up in theory that they are not impartial. Let's face it, none of us are sensor engineers, so we cannot separate theory for engineering advancements.

Click to expand...

No, I'm just not making assumptions based on the examples you posted above.

tomtom50 said:

Why deny the existence of a tradeoff Canon and Nikon engineers clearly saw?

Click to expand...

Again you're assuming you see a tradeoff they saw too, ignoring a lot of other possibilities, speed and cost not being the least (Canon historically also had more trouble to overcome with higher pixeldensities in general due to their different ADC layout, which in part explains why they seem to have hit a brick wall).

tomtom50 said:

I have actually talked with the designer of a cmos sensor used in high speed video cameras (Vision Research Phantom), and he emphatically told me that bigger pixels help DR and sensitivity, ALL ELSE HELD EQUAL. I suspect he knows more than any of us.

Click to expand...

That's all great, but Sony Semi conductors claims there is no such practical tradeoff anymore, atleast not for their designs. I believe what I see and the current cameras are quite telling in this regard.

tomtom50 said:

TrojMacReady said:

The RAW files from the F31fd were fine, but not above what you would expect from the sensor size in terms of noise.

Click to expand...

I could say look at the reviews, which strongly contradict you. Instead I will note that the Fuji f31fd did not produce RAW files.

Ahem.

Click to expand...

Interesting point you make there, since you started off with assumptions about sensor performance, using the F31fd as an example.

So what exactly do these reviews tell us about RAW performance and where can we find them?

Luckily the Fuji S6000/S6500 had the same sensor and did give us RAW. Ahem.

TrojMacReady said:

tomtom50 said:

TrojMacReady said:

The RAW files from the F31fd were fine, but not above what you would expect from the sensor size in terms of noise.

Click to expand...

I could say look at the reviews, which strongly contradict you. Instead I will note that the Fuji f31fd did not produce RAW files.

Ahem.

Click to expand...

Interesting point you make there, since you started off with assumptions about sensor performance, using the F31fd as an example.

So what exactly do these reviews tell us about RAW performance and where can we find them?

Luckily the Fuji S6000/S6500 had the same sensor and did give us RAW. Ahem.

Click to expand...

The DPR S6000 review says:

"Class-leading high ISO performance; might not be fantastic, but it's the best you'll get "

"Just like the F30, the S6500fd's output at ISO 800 is considerably better than most of its competitors manage at ISO 400 - and some at ISO 200. To have a small-sensor camera capable of producing results that are perfectly usable at ISO 800 is a luxury we have rarely seen before, and something for which Fuji must be congratulated."

Maybe you saw nothing special, others did.

May the force be with you
--
Alan.

Sorry. Things were getting a little nasty in the thread so this was my attempt to lighten up and take a deep breath.

B&H has their 128gb 133x Class 10 cards listed at $89.95, or 70 cents per gb. Probably only while the supply lasts.

BertIverson said:

javafool said:

...The true reason for a 20.1 megapixiel sensor in a point and shoot camera must be a conspiracy funded by the memory card suppliers...

Click to expand...

I doubt that. SD cards are about $1 / GB these days and they are, of course, reusable. So about 800 raw fit on a $16 card or about 3000 JPG.
Bert

Click to expand...

--
E-620 & E-30 DSLRs, E-P3 & OM-D E-M5, Canon S95 & SD4000 P&S

TrojMacReady said:

flektogon said:

wb2trf said:

Put it in the headline: lowering pixel count does not reduce noise or improve IQ (at any viewing distance for equivalent size images and same wafer fab capabilities for the chip).

Click to expand...

Yes, but on the other side increasing the pixel counts doesn't improve the IQ either. Just look at the reviews of the latest Canon and Nikon FF cameras. One has 24 Mp sensor and the other one has 38 (or 36 only?) Mp, but the reviewers give a nod to Canon. So, it looks like there exists an optimal pixel count (and the optimal pixel size), just it is not clear which one it is.

Click to expand...

Dpreview didn't give the nod to Canon. At high ISO they are about equal, at low ISO the Nikon shows better detail (and much better DR). No losses (other than file size and a few fps).

Click to expand...

I am not an expert on the digital cameras but I know a lot about the digital signal processing, which is ruled by the sampling theory. This theory says that the sampling frequency should be 2 times higher than the highest frequency component in the sampled signal. In the digital imaging this means that there should be exactly 2 pixels for the highest lpm (line pair per millimeter) resolution of the lens. As I remember, the best SLR lenses delivered up to 90 lpm of resolution, which on the 24 x 36 mm film field would yield 2160 x 3240 line pairs. Let’s say that today’s lenses have (slightly) higher resolution then a full frame sensor could have 4800 x 7200 pixels, or 34 Mp. So, yes, Nikon D800 uses such an “optimal” sensor. But what about the small sensor cameras including RX100? A 20 Mp sensor contains something like 3600 x 5500 pixels. To utilize fully such high sensor density, the lens should deliver something like 180 – 200 lpm. Does the CZ lens of the RX100 deliver such resolution? Maybe yes, as those “micro” lenses have higher resolution than the 35mm lenses. But what about the latest Sony P&S cameras with the 1/2.3” sensors having 18 Mp? To utilize such enormous pixel density the lens should deliver 350 lpm. Is this possible, that those miniature lenses can be so good? I personally doubt, but yes, the Sony RX100 might be equipped with an optimal sensor (at least from the resolution point of view).

I don't know what various outputs the RX100 offers, but, say, I had the options of 20 MP and 10MP out of cam.

Now in regard to IQ (resolution, noise etc.) would you suggest to shoot in default 20MP and downscale afterwards in Photoshop to 10MP, or is in-camera downscaling (or –sampling?) preferable – or is there no difference? The Nokia PureView 808 has ca. 38 MP and downsamples, so I guess it s a different story here. However, I must admit that I downloaded a 38MP and a 5MP downsampled image with the same motif and downscaled the 38MP to 5MP in Photoshop – no difference to the downscaled image (all lowest base ISO 60).

here s a little comparison that shows the downsampling advantages (probably only relevant in high ISO?):

http://asia.cnet.com/shootout-nokia-808-pureview-vs-panasonic-lumix-dmc-lx5-62216561.htm

tomtom50 said:

TrojMacReady said:

tomtom50 said:

TrojMacReady said:

The RAW files from the F31fd were fine, but not above what you would expect from the sensor size in terms of noise.

Click to expand...

I could say look at the reviews, which strongly contradict you. Instead I will note that the Fuji f31fd did not produce RAW files.

Ahem.

Click to expand...

Interesting point you make there, since you started off with assumptions about sensor performance, using the F31fd as an example.

So what exactly do these reviews tell us about RAW performance and where can we find them?

Luckily the Fuji S6000/S6500 had the same sensor and did give us RAW. Ahem.

Click to expand...

The DPR S6000 review says:

"Class-leading high ISO performance; might not be fantastic, but it's the best you'll get "

Click to expand...

You failed to answer the question and came up with a meaningless quote.

You're still talking (as they were) about jpegs. They didn't compare RAW. And possibly even worse, it has an 68% larger sensor than the cameras they compared it to (S3, H2, FZ7).
They did make some notes when examining the RAW mode:

"Here, with both Luminance and Chroma noise reduction their defaults, you can see just how noisy the output from the chip is, and just how much work Fuji is doing on the files to produce the final image"

tomtom50 said:

"Just like the F30, the S6500fd's output at ISO 800 is considerably better than most of its competitors manage at ISO 400 - and some at ISO 200. To have a small-sensor camera capable of producing results that are perfectly usable at ISO 800 is a luxury we have rarely seen before, and something for which Fuji must be congratulated."

Click to expand...

See above. Your assumptions are getting your theory nowhere.

tomtom50 said:

Maybe you saw nothing special, others did.

Click to expand...

Those that examined (and some measured) the noise in the RAW files, didn't. The jpeg engine was the special one.

Hard to tell but an FZ150 with a pixeldensity that is 2.5 times higher than the RX100 and having a compromised ultrazoom lens, still resolves more in RAW than my FZ18, which already beat the pixeldensity of the RX100 by around 50%.

Thanks for clipping out parts where I suggest disadvantages of more pixels.

I've never seen any proof of any optical advantage in raising sensor pixel account beyond 10 or 16. The only obvious "advantage" is shelf appeal to consumers to rank cameras by pixels / dollar. Perhaps a medium format, or even a FF, camera could support exceptional pixel count to some advantage, but be irrelevant to anything but perhaps aerial spying.

The Canon GF10, perhaps the best consumer videocamera, uses a sensor with only 2mp, and its low light results are unsurpassed. Unfortunately, we are unlikely to see side by side comparisons with video shot with an RX100 or even a NEX 5n, but the low light scores earned by the GF10 surpassed those of the NEX 5n in a camcorderinfo analysis, and the same may be true (must wait to see) relative to low light video shot with the RX100.

NowHearThis said:

meh, a GF10 doesn't fit in my pocket.

Click to expand...

An RX100 with a 12MP 1" sensor would, and also allow for cropping. One with a 5MP sensor would also fit in your pocket and still shoot good full resolution pictures and perhaps better video (fewer pixels: less aliasing or moire).

But the point is this: if the GF10 2MP sensor sweeps the field in terms of low light video (requires less than a Mii !), according to the camcorderinfo review, and also attains sharper resolution and color balance at low light, then one can argue that fewer pixels are better.

Your fundamental argument is, though, that "whatever is, is right," and that if companies makes something a particular way, it must be for the best or highest reasons. This is a very strange idolatry. Yet, you are at least half right: no company ever lost money by appealing to base illusions and fancies. But not everyone is a glutton for humbug.

flektogon said:

TrojMacReady said:

flektogon said:

wb2trf said:

Put it in the headline: lowering pixel count does not reduce noise or improve IQ (at any viewing distance for equivalent size images and same wafer fab capabilities for the chip).

Click to expand...

Yes, but on the other side increasing the pixel counts doesn't improve the IQ either. Just look at the reviews of the latest Canon and Nikon FF cameras. One has 24 Mp sensor and the other one has 38 (or 36 only?) Mp, but the reviewers give a nod to Canon. So, it looks like there exists an optimal pixel count (and the optimal pixel size), just it is not clear which one it is.

Click to expand...

Dpreview didn't give the nod to Canon. At high ISO they are about equal, at low ISO the Nikon shows better detail (and much better DR). No losses (other than file size and a few fps).

Click to expand...

I am not an expert on the digital cameras but I know a lot about the digital signal processing, which is ruled by the sampling theory. This theory says that the sampling frequency should be 2 times higher than the highest frequency component in the sampled signal. In the digital imaging this means that there should be exactly 2 pixels for the highest lpm (line pair per millimeter) resolution of the lens. As I remember, the best SLR lenses delivered up to 90 lpm of resolution, which on the 24 x 36 mm film field would yield 2160 x 3240 line pairs. Let’s say that today’s lenses have (slightly) higher resolution then a full frame sensor could have 4800 x 7200 pixels, or 34 Mp. So, yes, Nikon D800 uses such an “optimal” sensor. But what about the small sensor cameras including RX100? A 20 Mp sensor contains something like 3600 x 5500 pixels. To utilize fully such high sensor density, the lens should deliver something like 180 – 200 lpm. Does the CZ lens of the RX100 deliver such resolution? Maybe yes, as those “micro” lenses have higher resolution than the 35mm lenses. But what about the latest Sony P&S cameras with the 1/2.3” sensors having 18 Mp? To utilize such enormous pixel density the lens should deliver 350 lpm. Is this possible, that those miniature lenses can be so good? I personally doubt, but yes, the Sony RX100 might be equipped with an optimal sensor (at least from the resolution point of view).

Click to expand...

Excellent post. On a technical level your math is correct for monochrome sensors. Bayer sensors get their resolution primarily from the green channel. The demoisaicing process 'blurs' resolution somewhat (a fact Foveon fans make much of). For this reason a Bayer sensor's resolution is about 30% lower than your calculations would show.

That minor adjustment aside, 1/2.3 high MP sensors are lens limited. 16MP APS-C sensors in the NEX 5n and other cameras ALWAYS show more detail than a 16MP P&S, at all ISOs.

P&S cameras typically do not have anti-aliasing filters. The lens cannot resolve high enough to exceed the sensor. The lens itself is the filter. This either is or is not sensible depending on your viewpoint. My understanding is that good anti-aliasing filters are not cheap. Perhaps from an engineering standpoint these cameras would ideally downsample to make files of appropriate size for the information in the image, but then they could not be advertised as 16MP.

You can see this in the Imaging Resource G1X and RX100 images. They are remarkably close; Sony did a great job. But the G1X low ISO images carry every bit as much detail despite the lower nominal resolution. Obviously 20MP is not carrying more information than 14MP, and Sony has more resolution than needed, probably for marketing reasons. The people on this site show the power of the 'more MP is better' meme. Why exactly is 20MP better than 14MP if you are not getting more detail? You are definitely getting bigger files!

Lenses are getting better and better for the reason you cite. Modern sensor are more demanding than film. A derided modern kit lens would be considered a top-rate lens 20 years ago, distortion aside. A big reason that FF works so well is that the lens does not have to be so good in lpm to get a lot of picture information, simply because you have more millimeters.

TrojMacReady said:

TrojMacReady said:

Maybe you saw nothing special, others did.

Click to expand...

Those that examined (and some measured) the noise in the RAW files, didn't. The jpeg engine was the special one.

Click to expand...

From the same review "What is strange is that the output is visibly different to the FinePix F30; and it's not as good, looking more processed and less detailed. This is particularly true of ISO 1600, and it seems to be related to the in-camera processing (the sensor is apparently the same) - processed raw files look a lot more like the F30's output."

"ISO 1600 / 3200 JPEGS not as good as F30 (can be solved by shooting raw)"

"throw in the fact that it shares the universally acclaimed 6.3MP Super CCD sensor found in the F30"

"On the downside the JPEG output shows fairly strong noise reduction artefacts at anything over ISO 200, something you will see if you try to stretch the 6 megapixels to larger (say 8x10 inch) prints, and the processing is - for my tastes - too harsh, too contrasty and too sharpened. The saving grace is the inclusion of a perfectly usable raw mode (if you can wait the four seconds or so between shots) - with a little practice you can tease some truly stunning results out of these files. I was surprised to see that the high ISO output isn't as good as the F30 unless you shoot raw - this must be the fault of the new Real Photo Processor II. "

The review exactly contradicts your statements.

I have noticed that bringing evidence to bear never makes a difference.

tomtom50 said:

TrojMacReady said:

tomtom50 said:

Maybe you saw nothing special, others did.

Click to expand...

Those that examined (and some measured) the noise in the RAW files, didn't. The jpeg engine was the special one.

Click to expand...

From the same review "What is strange is that the output is visibly different to the FinePix F30; and it's not as good, looking more processed and less detailed. This is particularly true of ISO 1600, and it seems to be related to the in-camera processing (the sensor is apparently the same) - processed raw files look a lot more like the F30's output."

"ISO 1600 / 3200 JPEGS not as good as F30 (can be solved by shooting raw)"

"throw in the fact that it shares the universally acclaimed 6.3MP Super CCD sensor found in the F30"

"On the downside the JPEG output shows fairly strong noise reduction artefacts at anything over ISO 200, something you will see if you try to stretch the 6 megapixels to larger (say 8x10 inch) prints, and the processing is - for my tastes - too harsh, too contrasty and too sharpened. The saving grace is the inclusion of a perfectly usable raw mode (if you can wait the four seconds or so between shots) - with a little practice you can tease some truly stunning results out of these files. I was surprised to see that the high ISO output isn't as good as the F30 unless you shoot raw - this must be the fault of the new Real Photo Processor II. "

The review exactly contradicts your statements.

Click to expand...

It doesn't contradict anything. The above could have been quoted from just about any camera with a RAW capability. I quoted the review and noted they didn't compare RAW. The only direct comparison done is witin "its class" verus cameras with considerably smaller sensors.

So your example of the S6500/S6000 and specifically this review proves nothing about sensors and pixeldensities.

tomtom50 said:

I have noticed that bringing evidence to bear never makes a difference.

Click to expand...

You forgot to note that you brought zero evidence, which at this point apparently is hard to admit.

People like John Sheehy have done actual measurements, which I take over jpeg comparisons between cameras with differently sized sensors and RAW comments without a context or reference.

Here's a RAW example:
http://forums.dpreview.com/forums/read.asp?forum=1000&message=28927700

TrojMacReady said:

It doesn't contradict anything. The above could have been quoted from just about any camera with a RAW capability. I quoted the review and noted they didn't compare RAW. The only direct comparison done is witin "its class" verus cameras with considerably smaller sensors.

Click to expand...

You claimed the RAW files from the S6000 (same sensor as F30) were nothing special, that the magic was in the jpeg processor.

My quotes from DPR describe the sensor as

"universally acclaimed"

They say the jpeg processor on the S6000 underperforms that on the F30, but equal results can be had from RAW.

They say using RAW allows "stunning results" with the S6000.

You are right that this does not prove the original point regarding pixel densities; my original point was that high ISO IQ dropped as Fuji increased MP in the F series, the 8MP F40 being worse than the 6MP F30, and the 12MP F50 being worse than the F40. This directly addresses pixel densities.

The performance of the Fuji 6MP sensor was exceptional for its time (Chipworks dissected one to find its secret), you claimed the RAW was nothing special (a digression), I showed that DPR thought the RAW was quite special. You never addressed the basic point I made, which was that high ISO IQ dropped as Fuji increased pixel density in the F series.

DPR was wrong about one thing. "throw in the fact that it shares the universally acclaimed 6.3MP Super CCD sensor found in the F30"

TrojMacReady does not acclaim that sensor. Does that say more about DPR or TrojMacReady?

javafool said:

Sorry. Things were getting a little nasty in the thread so this was my attempt to lighten up and take a deep breath.

Click to expand...

Bert

Marti58 posted this link to a very interesting article about pixel size/density:

http://petavoxel.wordpress.com/2010/01/19/diffraction-fraud/

I personally think that the bigger the Mp the better until reaching at least the same Film resolution, but always accordingly to light physics as described in the article above and having the option to get smaller Mp/Mb files if desired.
--
aaanouel



Truth is a pathless land.
The dead past darkens the ever living present.

Corrections and critics are more than very welcome, desirables.