forpetessake: Sony is soon going to make Canon a third party lens manufacturer.
ttran88: "A $3200 camera will hardly give you a 1% increase in market share. Troll better"
True, but the tech in this will filter down to the cheaper cameras in the Sony range.
It's good to see that Canon will still continue as a going concern, and be able to play a niche role in the photographic industry, as the innovation in their DSLR range decreases :-)
Lee Jay: No AA filter = no sale.No OVF = no sale.I suspect I'll never own a Sony camera since they'll never produce something I'd buy.
Lee Jay: "And AA filters are so important that I wouldn't buy a camera without one unless it were in the gigapixel range."
You do release that the sensor pixel size + AA filter + lens are effectively a series-connected optical low-pass-filter? Just how much resolution do the lenses you use have? John Sheehy's calculations indicate that 150MP on a FF should be sufficient with even the finest lenses.
If you're so worried about moiré in these current "low-res" sensors without AA filters, there is a solution. Since Sony uses IBIS, they could have a switchable AA simulator if they license the tech from Ricoh/Pentax. Just involves oscillating the sensor up to 1px of rotary movement via the IBIS .
FocusBogus: Still not fully electronic shutters on mirrorless cameras in 2015? Sounds like a conspiracy: camera industry can sell some generations of cameras with mechanical shutters and later generations without them.
frzRom: "But BSI + on chip phase AF + on chip Dram ... Not sur it is possible to do it."
With stacked BSI tech (Exmor RS), pixel memory for a global shutter without any reduction in fill-factor may well be possible.
For more info on the problems involved with implementing a global shutter see:
Androole: I think the SuperRAW is obviously the most (I would argue really the only) significant element of this announcement.
However, it does make you feel like there's a bit of an agenda given that they carefully measured the performance of the SuperRAW in absolutely ideal test conditions, yet they decided that it wasn't worth it for them to measure the E-M5 II's High Res Mode when they reviewed it just 2 months ago. Hmmm...
And I bet they won't test the Pentax K-3 II's multi-exposure mode, either...
But DxO only deals with raw files. If the stacked mode is only available for JPEGs, then DxO won't measure it. PhotoAcute, and now DxO, offers the result of stacking in a DNG output format.
BTW, Pentax offers a multi-raw shot HDR file in PEF format, but this is really a container format than contains multiple separate raw shots, each with different exposure parameters, rather than one HDR shot with combined raw data.
The comparison of the noise of the D810 and the A7S in the 2nd page of the article was not a good choice, as the A7S is likely to be using Aptina's (now ON Semiconductors) DR-Pix technology.
There is a step in the read noise between ISO1600 and IS2000+, with the higher ISOs show less RN than expected. DR-Pix technology allows the changing of the gain in the sensor, so the sensor designer's trade-off between Full Well Capacity, Conversion Gain & Read Noise can be altered in-camera.
A better pixel-size comparison choice would have been between the Pentax K-5 (16MP) & K-3 (24MP), both APS-C cameras.
On the DR graph you can use the Screen tab to seen the pixel-level performance, while the Print tab shows the result using the same output size e.g. same print size.
coyot3: I tested on my nikon d3100 and the results are at iso 100 1" f4 underexposing 4 ev and later recovering in Darktable(linux) give less noise than iso 3200 1/30 f4 no exposure changes.
I dont know if i understand right the article xD im going to re read ._.
¿Some thoughts on my results?
You'll find an ISO100 + 4-stops boost VS ISO1600, using a K-5, here:
The exposure and sensor was the same in both cases: same scene luminance, same shutter speed & same f-stop. So the shots have the same amount of shot noise. But the ISO100 image was sent to the ADC at 1/16th of the level of the ISO1600 shot, so it was "under-brightened" in the stored image file.
DuncanDovovan: Richard / Rishi: What I would really love to see is a shot noise simulation. Rishi looks to be the guy who can calculate the number of photons that a pixel gets.
From the number of photons caught, couldn't you calculate the S/R ratio?
And from the S/N ratio, couldn't you visualise/simulate the shot noise how it would look like in a 100% crop (pixel level) and a crop that would show the scene at the same scale (more pixels scaled down to the same magnification)?
Wouldn't it be possible to visualise the shot noise, other upstream noise and downstream noise separately that way?
I would love to see (as in picture crops / not tables) and compare the shot noise, upstream and downstream noise components under several conditions. Like higher vs lower ISO. Shorter vs longer exposures. Overexposure and compensating, underexposure and compensating vs normal exposure.
And all of that for 5 different tonal values from dark to bright and in different situations from cloudy to bright daylight.
"I would expect the bright areas to have less noise."
No, the bright areas have more shot noise because they are the result of more photons hitting the sensor in this area. (Shot noise increases at the sq root of the number of photons.) It's just that the Signal-to-Noise Ratio will be better in these areas, so you probably won't notice the extra noise.
Two examples:A: 1,000,000 photons. B: 10,000 photons.
A. Shot noise = sq-root(1,000,000) = 1,000.SNR = 1,000,000/1,0000 = 1,000:1
B: Shot noise = sq-root(10,000) = 100.SNR= 10,000/100 = 100:1
BTW, SNR is sq-root(number of photons), but I've worked it out here to show the amount of the shot noise.
www.sensorgen.info has been mentioned a few times in these comments. It is a site run by Prof. Bob Newman, which shows the Full-Well Capacity of sensor pixels and Total Read Noise, as calculated from DxOMark measurements.
I'm interested in examining the Total Read Noise, input-referenced in -e (photo-electron charges), curve-fitted to estimate the Sensor read noise (upstream of the PGA) and the ADC noise (downstream).
Here's a sheet that shows a few cameras.
There are more cameras on the RN (Read Noise) Contributions tab, and it's easy to use it as a template to add more cameras.
SNR (dB) is my calculated value. You can compare it against SNR (DxO) to see how well the curve-fit performs. On the RN Contributions page you also see this in the "Diff square" value.
The 16x Ratio compares the Total RN at base ISO & 16x base ISO. The closer this is to 1, the more "ISOless" the camera.
Duncan, there is already a shot noise simulation in Wikipedia:
SolidMetal: Great article! Though I still cant really understand why high ISO images have less dynamic range. That orange line which is overamplified in the picture wouldnt be out of the raw file even if its straight not curved? I mean its still brighter than the highlights marked in yellow on the low ISO picture.
SolidMetal, 6 stops less DR if there was no DR reduction at low ISO due to the ADC contributing some noise. But the ADC contributes either a little noise (Sony, Nikon, Pentax) or moderate noise (Canon). This means that the straight-line relationship of 1 stop less DR for each stop increase in ISO is usually not reached until the mid or high ISO ranges.
See my DR graph and DR reduction explanation in:
That's the 2nd part of an explanation of the role of ISO Sensitivity control in a digital camera. The 1st part is here:
SolidMetal, there are two ways that hightlights get blown:'
1. Sensor pixels saturating (filling up to overflowing).2 ADC clipping (running out of bits in a 12-bit/14-bit digital number).
The gain stage between the sensor & the ADC is set up so, at base ISO, the sensor saturates just before/at the same time as the ADC clips. This gives max. DR.
With weaker & weaker exposures, less photons are captured, so the sensor pixels are "less well-filled". So once you start boosting the gain to compensate for weak exposures and make the rendered image brighter, it becomes less & less likely that any pixels will have saturated.
But with more & more gain, any relatively high pixel values are now more likely to clip the ADC. From ISO100->ISO1600 is 16x gain or 4 stops. If a pixel saturates at 48,000e- (photo-electrons) at ISO100 and the ADC also just clips at this level, then at ISO1600 the pixel will only reach 3,000e- before the ADC clips instead.
So the DR decreases with increasing ISO.
Muqdad: If RC of L.Rentals is so insistingly positive about Canon's response and reaction, and the general impression on Canon's customer service is almost of the very best out there among manufacturers, shouldn't an honest will to be helpful to Canon in this matter made him more willing to give them a supposedly deserved chance to show their good support action expected by R. Cicala himself?
Rather, he went trigger-happy and did this HUGE damage to the company we applaud for caring for and actually respecting their customers?
I can only think he believed he may lose the chance to be the one who first tells about it. This is exactly the perfect abuse of authority.
"The spots may be part of the fabrication, e.g. a film coating that is required to limit contact between the laminates, and once the laminates relax for a period after fabrication, the spots go away. So in that case this is not a problem"
Interesting conjecture. So the cameras come from the factory under-matured and should not be used for serious work until they've had time to age?
Haven't heard that one before. Isn't the internet wonderful?
I don't see it the way. If an individual has this problem they can get fobbed off with a line like "you must have done something to the camera" (e.g. stored it inside a hot car). But a rental company with a bigger sampling goes public about it so, if others have an issue, they're less likely to be fobbed off.
Manufacturers have to take the heat when their Quality Assurance process fails. Why should companies and individuals who buy their equipment worry about the manufacturer's reputation? That's a fanboi reaction.
The manufacturers spend enough telling us how wonderful their products are and how our lives will be better if we buy them. Some negative feedback from owners helps to rebalance this a little in the direction of reality.
D Bowcut: I have posted an open-sky shot in the Gallery at f22, ISO 150, from my new T6s: This is at 100%, but re-sized to fit the Galleries rqmt.
I can't tell if the problem exists here or if this reveals only expected pixelization? My camera s/n is 022031000385.
Otherwise, camera is running OK.
Lock up the mirror for cleaning and look at the top of the sensor with a loupe or magnifying glass.
This looks like a defect in the "toppings" (anti-refection layer/IR-filter/ perhaps even microlenses) which are often added by specialist firms. Don't know if Canon adds the toppings in-house or sends their sensors outside for these. (They make their own DSLR sensors.)
You wonder why this type of obvious defect wasn't picked up in the visual QA stage? Perhaps it's a chemical reaction that takes some time to form bubbles?
Earth Art: Thermal conditions make a huge difference in shot noise as well. Big difference shooting at night in 30 deg F and 80 deg F.
That's not shot noise. Rather it's dark current & thermal blooming if parts of the sensor (the columnar ADCs situated on the edge) are operated in a long enough exposure.
Shot Noise aka Photonic Noise is related solely to the number of photons in the capture. It's an intrinsic property of light, so it can't be avoided.
thk0: "There are three factors that affect how much light is available for your sensor to capture: your shutter speed, the size of your aperture (not f-number) and the size of your sensor."
Yes, f-number. If you use the physical aperture size, then the sensor size does not matter (assuming the same sensor coverage of the light circle). The amount of light passing through the lens, and hence "available for your sensor", only depends on the area of the entrance pupil, transmission efficiency and the duration of the exposure. If you use f-number, then the physical aperture scales with the sensor size and sensor size becomes part of the equation.
(edits for typos)
Another way of stating this: At the same exposure level and similar sensor technology, the shot noise SNR is proportional to the sq root of the sensor area. So 4x bigger sensor = 2x better shot noise SNR.
"There are three factors that affect how much light is available for your sensor to capture: your shutter speed, the size of your aperture (not f-number) and the size of your sensor."
The "Exposure Triangle" is Shutter Speed, F/number, Scene Luminance. That's why you use a flash or a light reflecting panel: to increase the scene luminance component of exposure. These 3 determine the exposure (the photonic density captured per mm2 during the exposure period). So if these 3 remain the same, with larger sensors you get the same exposure, but more Total Light (more photons) being captured. For the same exposure, there's more photons collected, so there's also more shot noise photons in a bigger sensor capture. But, since the Shot Noise SNR is proportional to the sq root of the number of photons, it's still better with a bigger sensor.
Example. Small sensor: 10,000 photons captured, SNR = 10,000/100= 100:1
2.3x bigger sensor: 23,000 photons captured: SNR = 23,000/151.7 = 151.7:1
dosdan: "we wish Adobe went a bit further and allowed for simple image averaging with Raw DNG output."
Photo Acute has had this ability for years (2007), as well as a lot more. The parent company, Almalence, has been around since 2005.
Here's a stacking test of 14 handheld K20D raw files (the max. I could do in a burst before the shot buffer filled up), each taken at ISO3200, from 2009: