5D Mark III/X Information [CR1]

[sublime LightWorks]

Where, pray tell, has it been "confirmed" all over the place that pixel size has nothing to do with ISO noise performance? 

Smaller pixels perform better at high-ISO all the way until read noise starts to dominate (which is way, way out there for most sensors).  These were shot at the same ISO, same shutter speed, same f-stop, same focal length, both use the same sensor area, both were shot in raw and processed in the same software.  The pixel area is different by a factor of 16.  The processed images on the far right column tell the story - the smaller pixels preserved more detail with less noise than the bigger pixels did even though they were set at their maximum ISO.

http://photos.imageevent.com/sipphoto/samplepictures/Pixel%20density%20test%20results.jpg

Ok, let's say this is correct (I'm not buying this as I can point to 1000 other examples of images taken where this is not the result).  Can you explain why comparisons of exactly the same RAW images taken with a 5Dmk2 at ISO 3200 are so much cleaner than the 7D?

Because the comparison isn't done correctly.  It's not done with the same sensor area used.  It's done with both at 100% or both full-frame.  Neither is correct if you're talking about pixel density.  If you change nothing but pixel density, you have to assume the same amount of sensor area is used.  If you do that, you'll find your 7D out-performs your 5DII.  The reason the 5DII is better overall is because it has more sensor area.

Incidentally, those images I posted were from the 5D and the Canon S3IS compact.  The compact won easily when the 5D didn't have the advantage of it's massively larger sensor.

I'll need some time to wrap my head around what you are saying....been a long day at the office.  Let me digest that, but if you can post a use-case as an illustration, that would help.

One question as a background item:  do you know what the SNR is per pixel on say the 5Dmk2 and the 7D?  My BSEE head says the smaller pixels (photodiodes) have more leakage (assuming similar generation sensors, not old stuff vs. new stuff) than larger ones and hence more noise.

[Lee Jay]

Where, pray tell, has it been "confirmed" all over the place that pixel size has nothing to do with ISO noise performance? 

Smaller pixels perform better at high-ISO all the way until read noise starts to dominate (which is way, way out there for most sensors).  These were shot at the same ISO, same shutter speed, same f-stop, same focal length, both use the same sensor area, both were shot in raw and processed in the same software.  The pixel area is different by a factor of 16.  The processed images on the far right column tell the story - the smaller pixels preserved more detail with less noise than the bigger pixels did even though they were set at their maximum ISO.

http://photos.imageevent.com/sipphoto/samplepictures/Pixel%20density%20test%20results.jpg

Ok, let's say this is correct (I'm not buying this as I can point to 1000 other examples of images taken where this is not the result).  Can you explain why comparisons of exactly the same RAW images taken with a 5Dmk2 at ISO 3200 are so much cleaner than the 7D?

Because the comparison isn't done correctly.  It's not done with the same sensor area used.  It's done with both at 100% or both full-frame.  Neither is correct if you're talking about pixel density.  If you change nothing but pixel density, you have to assume the same amount of sensor area is used.  If you do that, you'll find your 7D out-performs your 5DII.  The reason the 5DII is better overall is because it has more sensor area.

Incidentally, those images I posted were from the 5D and the Canon S3IS compact.  The compact won easily when the 5D didn't have the advantage of it's massively larger sensor.

I'll need some time to wrap my head around what you are saying....been a long day at the office.  Let me digest that, but if you can post a use-case as an illustration, that would help.

Just what we're talking about - 1Dx versus D800.  Same sensor size, different pixel size.

One question as a background item:  do you know what the SNR is per pixel on say the 5Dmk2 and the 7D?  My BSEE head says the smaller pixels (photodiodes) have more leakage (assuming similar generation sensors, not old stuff vs. new stuff) than larger ones and hence more noise.

My MSEE head says leakage is a very, very minor issue on the very large pixels (and not much related to noise anyway).  Remember, the state-of-the-art in pixel design is for cell phone cameras - 1-1.5 microns.  4-7 micron pixels like we use in our dSLRs are enormous by comparison.

Since you're an engineer, I can explain very easily why smaller pixels are better, as long as read noise doesn't dominate.

What do large pixels do compared to small ones?  Spacial block averaging.  Now, which do you think is more effective at preserving detail and removing noise, spacial block averaging or modern noise reduction algorithms?  The answer is obvious - no one could sell any noise reduction software if it couldn't beat the daylights out of a simple block average approach, which is close to the lousiest filter you can imagine!

[Radiating]

Where, pray tell, has it been "confirmed" all over the place that pixel size has nothing to do with ISO noise performance? 

Smaller pixels perform better at high-ISO all the way until read noise starts to dominate (which is way, way out there for most sensors).  These were shot at the same ISO, same shutter speed, same f-stop, same focal length, both use the same sensor area, both were shot in raw and processed in the same software.  The pixel area is different by a factor of 16.  The processed images on the far right column tell the story - the smaller pixels preserved more detail with less noise than the bigger pixels did even though they were set at their maximum ISO.

http://photos.imageevent.com/sipphoto/samplepictures/Pixel%20density%20test%20results.jpg

That test is extremely poorly done.

The reality of the situation is that generally the more pixels you have, the more noise a camera will have. Not because the pixels somehow cause the noise, but because it reduces the light gathering ability of the camera's sensor. To actually put more pixels on a sensor you have to make the pixel wells themselves smaller. It's like dividing a house up into rooms. A house with just one big room has a lot more space than a house with walls everywhere. On top of that the current generation of sensors has the wiring for the pixels on the front of the sensor, meaning the more pixels you have the more wiring you have which isn't capturing photons. Now between old low megapixel sensors and new high megapixel ones there have been a lot of advancements. For example micro lenses have been developed which basically help to bend the light around all the non-light gathering real estate of a sensor and focus it onto the light gathering real estate. Also technology has been devloped to make sure the photons are more accurately read.

So in general assuming everything is equal more pixels means more noise. The test you posted obviously has the variables skewed to show otherwise likely by comparing a body which isn't very advanced to one that is. Furthermore most camera manufacturers lie about their iso settings to the point where they are inflated by 80% in many cases. So the camera is shooting at iso 900 but it says iso 1600 on the display. So comparing cameras outside a lab setting isn't very useful.

Anyways I encourage people to read about the quantum efficiency of a sensor:

http://www.sensorgen.info/

Often times a camera will go up from on generation to another both in reducing noise and increasing resolution, like the 1D III to the 1D IV through superior sensor design. Other times you have cases like the D3s and D3x where the D3x has twice as much noise and twice as much resolution than the D3s.

[LetTheRightLensIn]

Ok, let's say this is correct (I'm not buying this as I can point to 1000 other examples of images taken where this is not the result).  Can you explain why comparisons of exactly the same RAW images taken with a 5Dmk2 at ISO 3200 are so much cleaner than the 7D?  The 7D has a far smaller pixel, but tests done on many sites comparing it to the 5Dmk2 clearly show the degraded IQ on the 7D at the same high ISO levels.  And the 3Mpix difference between the 5Dmk2 and the 7D is not enough to account for that.

Because he was talking about a normalized comparison and you were talking about a 100% view comparison.
Per area of sensor the 7D actually seems to have a tiny bit LESS noise than the 5D2, however, the 5D2 has a LOT more area. So long as you are willing to shoot wide open and this with the same ISO, aperture and shutter speed and can frame the shoot as you desire with the lens you have and don't need to crop way in onto a distant bird or something, the 5D2 result is better. But if you are distance limited then you end up comparing based on how well they do per area of sensor if you want to compare them fairly or by 100% view if you want to compare detail and in either case the 7D actually beats the 5D2 (although it's not by any amount to really care about when it comes to the noise).

DPReview has examples of this in their 7D review and analysis.  And since I own both the 5Dmk2 and the 7D, I can tell you in the real world, the 7D cannot hold a candle to the 5Dmk2 at ISO 1600 or above.

It can if you are distance limited. And the fair comparison here, in the general sense, would be to imagine the 7D sensor not being clipped off at APS-C size but extended to FF size, then with it's slightly better performance per area and equal area you'd see that it's actually doing a trace better despite having much, much smaller photosites. It does have newer technology, but even if it did not it probably would not end up looking more than maybe 1/2 stop worse at worst.

In rare cases it can even look better at ISO3200 even in a non-distance limited case if the scene consists mostly of deep shadows since it has less banding there. But usually, when you can frame as you desire and if oyu use the same exposure, yeah the 5D2 will look better than the 7D, but it's really because the 5D2 has an amazing 2.56x the surface area to collect light over due to the much larger sensor.

[LetTheRightLensIn]

My only pick at that is the images have been re-sized, which renders them with an averaging of the noise, not a true noise image.  If I am shooting a 36Mpix camera and I have to resize the image to get it to look as good as my 5Dmk2, then what am I getting for 36Mpix?

re-sized/normalized is the only fair way to compare them

you are buying you can toss away all the extra detail and end up with the same detail and yet not any worse noise for the times you care about noise the most AND yet you can also have other options that you won't have with the 5D2:
1. sometimes you may value detail above all else and just live with both more detail and more noise and prefer that to less noise and less detail
2. if you shoot at low ISO the amount of extra noise won't really matter since barely any noise to a trace more than barely any noise is still more or less barely any noise and yet you can get 50% more detail information in your photo
3. and just to repeat it again, if you are in a case where noise matter above all else, then you can filter away detail and end up with the same noise (and you could actually use special NR software and so on instead of a simple re-size and do better than expected and even make up for anything the tinier pixels lost and maybe even more)

I agree with you, a RAW, one-to-one comparison under identical conditions is whats needed.  Until that happens, we will not know.

Agree, we really don't much yet other than the detail in a few of the D800 photos is cool.

[LetTheRightLensIn]

Seeing as we've transitioned into talking about the D800, can somebody explain to me what exactly "uncompressed HD signal out" means?

I've seen people talking about this all over, but nobody explains what it actually is and what it is useful for. As far as I'm aware, signal sent through HDMI is already uncompressed. So what is the all the fuss about, or is it a one of those features that nobody has any idea what it is and just brags about it?

For someone with the equipment to handle it it means you avoid some cheap on the fly compression adding posterization, blocking, robbing details, etc.  Granted for the avg guy on the street it's somewhat just bragging rights. Although if you have the right device you might be able to grab it onto a laptop in the field and if you stick to short clips and then re-compress might not overflow your HD.

It could be nice for avg guy, at times, but generally it would be much more of a serious pro thing.

(I am assuming they are sending the video, as captured, out over HDMI during live filming, all the frames, non-skipped and before the internal compressor compresses it into h.264 or whatnot. Hopefully that is not a false assumption.)

And it would allow for better live monitoring, it won't be down-rezzed, or have frames skipped, or have had anything meddlesome done. I think the current DSLRS all did some combinations of those things during live previews. I have to say I haven't tried live preview of HDMI out yet so I'm not sure exactly what they are doing. I do know that you certainly can't get 1920x1080p full frame rate all frames with zero compression out of any of the Canons.

[LetTheRightLensIn]

In the past I believe the AF was either controlled by the main CPU (which generally hadn't been the digics, but an additional chip AFAIK) or an additional AF only CPU.

On the 1DX they will use a Digic iv as the AF chip.

Most of the rumors have mentioned single digic V for image throughput and some have also mentioned digic iv for AF. Who knows if any of this stuff is for real, the 22MP or 7fps either.

Anyway I don't think 3 digics, if it came to that, shoud be a shock in the general sense (coming from Canon, yes, perhaps). How much do you think a digic iv costs? they use them in the lowest of lowest priced Canon P&S cameras. And, I forget, but I thought they added a separate CPU for AF even in the 7D, although not a digic chip.

Why is it so insane to think (other than because of the sort of company Canon has become)? Other companies do it. If you think it is all insane, in principle (in reality, perhaps you are correct), you've had too much of the Canon kool-aid.

Definitely not drinking the Canon kool-aid, far from it. 

Let's drive on that idea of 3 processors.  Could all that fit in a 5D body?  I'm thinking no, but.....it's possible.  Looking at the circuit board images of the 1Dx, I have a hard time thinking that hardware is going to fit in a 5D body.  Gotta remember, that 1Dx has the advantage of the room offered by the built-in grip.

In the 7D it uses the dual Digic's for the AF as well....check the specs and the circuit board layout...nothing indicates another AF processor in addition to the Digics.

I'll have to check but I could swear they said they added a little chip for 7D AF.

Well the 7D proves they can fit two digics and an additional master cpu and maybe a small additional AF CPU.

However, the digic chips do seem to take up more surface area than any of the other chips. Is there room to fit in three digic chips? Not sure. I'd have to look at the boards and think. It's possible you could have a point there if there is nothing else they can shrink or move around.

OK, let me see, well a quick bit of google quickly turns up lots of references to a dedicated AF chip being used in the 7D. So I think it has, but I think this chip must be noticeably smaller than a digic sin emy vague recollection was a circuit board with two big digics and then a bunch of much small chips, if I recall correctly (and I might not be).

I managed to find a very, very tiny pic of the 7D mainboard, it's so small that it is hard to say, it looks like they might just be able to cram in another digic, but it depends. Image it too small to really tell and perhaps the Digic 5 is much larger, etc.

[Lee Jay]

Where, pray tell, has it been "confirmed" all over the place that pixel size has nothing to do with ISO noise performance? 

Smaller pixels perform better at high-ISO all the way until read noise starts to dominate (which is way, way out there for most sensors).  These were shot at the same ISO, same shutter speed, same f-stop, same focal length, both use the same sensor area, both were shot in raw and processed in the same software.  The pixel area is different by a factor of 16.  The processed images on the far right column tell the story - the smaller pixels preserved more detail with less noise than the bigger pixels did even though they were set at their maximum ISO.

http://photos.imageevent.com/sipphoto/samplepictures/Pixel%20density%20test%20results.jpg

That test is extremely poorly done.

And yet, you have nothing to say what about it was poor except that the answer isn't what you expected it to be.  Two cameras, same generation, same f-stop, shutter speed, focal length, ISO, light level, sensor size, capture method and processing.  The only significant difference is the pixel size, and it's different by such a large amount (factor of 16) that it swamps out other minor effects.

The reality of the situation is that generally the more pixels you have, the more noise a camera will have. Not because the pixels somehow cause the noise, but because it reduces the light gathering ability of the camera's sensor. To actually put more pixels on a sensor you have to make the pixel wells themselves smaller.

And there's more of them.  The overall effective fill factor is about the same between the 5D and the G12.  So I wouldn't hang the hat of your argument on that issue.

It's like dividing a house up into rooms. A house with just one big room has a lot more space than a house with walls everywhere. On top of that the current generation of sensors has the wiring for the pixels on the front of the sensor, meaning the more pixels you have the more wiring you have which isn't capturing photons.

House rooms don't have microlenses.

So in general assuming everything is equal more pixels means more noise.

No.  It means the pixels have more noise, and the sensor has the same photon capture with more detail.

The test you posted obviously has the variables skewed to show otherwise likely by comparing a body which isn't very advanced to one that is. Furthermore most camera manufacturers lie about their iso settings to the point where they are inflated by 80% in many cases. So the camera is shooting at iso 900 but it says iso 1600 on the display. So comparing cameras outside a lab setting isn't very useful.

None of that is relevant because f-stop, shutter speed and focal length were the same.  Thus, the number of photons that struck the sensor was the same.  ISO calibration is irrelevant.

Anyways I encourage people to read about the quantum efficiency of a sensor:

http://www.sensorgen.info/

Often times a camera will go up from on generation to another both in reducing noise and increasing resolution, like the 1D III to the 1D IV through superior sensor design. Other times you have cases like the D3s and D3x where the D3x has twice as much noise and twice as much resolution than the D3s.

The sensors I tested were of the same generation and of about the same QE.

[x-vision]

http://photos.imageevent.com/sipphoto/samplepictures/Pixel%20density%20test%20results.jpg

Sorry but this is an apples vs oranges comparison.

Due to higher production volumes and economies of scale, the small sensors in digicams and cell phones are made on newer and more advanced technologies than the larger sensors in DSLRs.

So, it’s not the small pixel size that makes image quality better but the newer, more advanced tech used in the small sensors.

For the same technology, the house wall analogy applies: a larger number of rooms leads to increased wall area and decreased living space (for any given area).

And no, no microlens magic can (fully) compensate for loss of active area on a sensor.

[Radiating]

The sensors I tested were of the same generation and of about the same QE.

That's not true at all. The G12 has an extremely advanced sensor that has a QE of 55% the 5D has a QE of 25%. That's more than a stop of difference. You're comparing stone age technology to the space shuttle.

I think the problem with your test is that you have to account for the number of t/stops in each camera's respective lens. If you actually shot the images above at the same f/stop, shutter speed and iso and didn't just try to get the exposure the same, then most likley the lens you were using transmitted twice as much light onto your 5D as the lens on the G12 due to the quality of the glass.

In any case. The point is that while it is possible to make a 300 megapixel monster of a camera with 55% quantum efficiency, they haven't done so. My guess is that the cost of scaling up the G12 sensor makes it not cost effective.

I'm positive that Canon has the technology to make a camera that has 36 mp and performs better than the D3s in low light and dynamic range. Unfortunately what is most likley going on is that they realize that such a camera would end up costing more more than creating 2 seperate bodies, one for resolution and one for low light.

[simonxu11]

The sensors I tested were of the same generation and of about the same QE.

That's not true at all. The G12 has an extremely advanced sensor that has a QE of 55% the 5D has a QE of 25%. That's more than a stop of difference. You're comparing stone age technology to the space shuttle.

I think the problem with your test is that you have to account for the number of t/stops in each camera's respective lens. If you actually shot the images above at the same f/stop, shutter speed and iso and didn't just try to get the exposure the same, then most likley the lens you were using transmitted twice as much light onto your 5D as the lens on the G12 due to the quality of the glass.

In any case. The point is that while it is possible to make a 300 megapixel monster of a camera with 55% quantum efficiency, they haven't done so. My guess is that the cost of scaling up the G12 sensor makes it not cost effective.

I'm positive that Canon has the technology to make a camera that has 36 mp and performs better than the D3s in low light and dynamic range. Unfortunately what is most likley going on is that they realize that such a camera would end up costing more more than creating 2 seperate bodies, one for resolution and one for low light.

It actually doesn't matter whether Canon has that kind of technology or not, nobody denies that Canon has excellent R&D in this industry.

It just doesn't mean anything to end-users if Canon don't put the technology in their products. Someone mentioned Canon is a conservative company which I totally agree.

Canon can still maintain their sales figure ahead anybody else just because there's no company like Apple in this industry.

IMO, Nikon focus on how to make better DSLRs and Canon focus on how to sell more DSLRs .

[D.Sim]

The sensors I tested were of the same generation and of about the same QE.

That's not true at all. The G12 has an extremely advanced sensor that has a QE of 55% the 5D has a QE of 25%. That's more than a stop of difference. You're comparing stone age technology to the space shuttle.

I think the problem with your test is that you have to account for the number of t/stops in each camera's respective lens. If you actually shot the images above at the same f/stop, shutter speed and iso and didn't just try to get the exposure the same, then most likley the lens you were using transmitted twice as much light onto your 5D as the lens on the G12 due to the quality of the glass.

In any case. The point is that while it is possible to make a 300 megapixel monster of a camera with 55% quantum efficiency, they haven't done so. My guess is that the cost of scaling up the G12 sensor makes it not cost effective.

I'm positive that Canon has the technology to make a camera that has 36 mp and performs better than the D3s in low light and dynamic range. Unfortunately what is most likley going on is that they realize that such a camera would end up costing more more than creating 2 seperate bodies, one for resolution and one for low light.

It actually doesn't matter whether Canon has that kind of technology or not, nobody denies that Canon has excellent R&D in this industry.

It just doesn't mean anything to end-users if Canon don't put the technology in their products. Someone mentioned Canon is a conservative company which I totally agree.

Canon can still maintain their sales figure ahead anybody else just because there's no company like Apple in this industry.

IMO, Nikon focus on how to make better DSLRs and Canon focus on how to sell more DSLRs .

You can't sell more cameras with making them better. Their sales are ahead because they're marketed well, and their product is good.

[torger]

When I see Canon products coming out today I get the impression that pretty much everything except the sensors are great. I'm very impressed with the lenses coming out for example. However, to some of us the sensor is quite an important part of a camera. When it was D700 vs 5Dmk2 it was not a hard choice, about the same performance plus a lot more resolution.

Then Nikon started to impress with sensors, D3s, D3x, D7000 and now D800. This is really tough competition concerning dynamic range, low light capability and now resolution. Canon is now lagging behind. Will 1DX or 5Dmk3 change that? I doubt it, but we'll see once the products are out and properly tested.

[D.Sim]

When I see Canon products coming out today I get the impression that pretty much everything except the sensors are great. I'm very impressed with the lenses coming out for example. However, to some of us the sensor is quite an important part of a camera. When it was D700 vs 5Dmk2 it was not a hard choice, about the same performance plus a lot more resolution.

Then Nikon started to impress with sensors, D3s, D3x, D7000 and now D800. This is really tough competition concerning dynamic range, low light capability and now resolution. Canon is now lagging behind. Will 1DX or 5Dmk3 change that? I doubt it, but we'll see once the products are out and properly tested.

the dynamic range on the nikons isnt so much the sensor, but more their active-d lighting. Canon has their HTP, which isn't as good IMo, but if you shoot RAW....

That and since Nikon are using Sony sensors =P



that said, it might seem that Canon are lagging behind slightly - but the product is still brilliant - and if you were to threaten to jump ship because of that you're failing to see the big picture, or have too much cash.  If Nikon *did* somehow win this round - and they haven't because Canon hasn't even revealed their cards yet - theres still no reason to jump ship, Canon will come right back, because thats the way it is.

[torger]

the dynamic range on the nikons isnt so much the sensor, but more their active-d lighting. Canon has their HTP, which isn't as good IMo, but if you shoot RAW....

That and since Nikon are using Sony sensors =P

that said, it might seem that Canon are lagging behind slightly - but the product is still brilliant - and if you were to threaten to jump ship because of that you're failing to see the big picture, or have too much cash.  If Nikon *did* somehow win this round - and they haven't because Canon hasn't even revealed their cards yet - theres still no reason to jump ship, Canon will come right back, because thats the way it is.

Yes on-chip ADC of Sony sensors seems to be key for DR. I hope they don't have a patent for it :-). I only consider RAW, so I'd say the practical DR is due to the Sony Exmor and the fact that they don't have as much pattern noise which subjectively kills some of Canon's usable DR.

Since I'm in the low ISO high resolution niche I'm a bit worried though, Canon seems to be that kind of player that care more about maximizing profit than to than to care for for all their users, including the less profitable niche users. But as you say, Canon hasn't revealed their cards yet. We'll see.

I would not "jump ship" though before we start to see what Canon will and can deliver. The latest three lenses suggest an interest for video (IS on wide angles) but also for high resolution (the 24-70 without IS but great MTF).