Sony to unveil a 50 MP new A7 body at Photokina?

[Lawliet]

And another person that couldn't photograph the inside of a paper bag

Well, guess I prefer sujets & subjects that actually make the cut for paid for publication.
Should the paper bag ever be a topic I'll defer to your expertise.

 

[RLPhoto]

And another person that couldn't photograph the inside of a paper bag

Well, guess I prefer sujets & subjects that actually make the cut for paid for publication.
Should the paper bag ever be a topic I'll defer to your expertise.

Oh really? By all means let us see you work on these publications. We're all dying to see them now that you mentioned them.

 

[psolberg]

And another person that couldn't photograph the inside of a paper bag

Well, guess I prefer sujets & subjects that actually make the cut for paid for publication.
Should the paper bag ever be a topic I'll defer to your expertise.

Oh really? By all means let us see you work on these publications. We're all dying to see them now that you mentioned them.

OMG dude, seriously? clearly high resolution photography and oversampling discussions are not something you're handling well. Can you just let us adults have a thread to discuss without this nonense? Relax and go enjoy whatever equipment you like.

 

[dtaylor]

I would argue that 22 MP vs. 36 MP is negligible with any lens. You need a 50% or greater gain on each axis before it really becomes evident in print.

having shot both, it isn't. But it depends on what you're doing. If you're web publishing at full HD or approximate, then yes.

I was printing samples to an Epson 3880, some crops scaled to appear as if they were being printed on a larger printer (i.e. up to 36" eq).

I'm open to evidence that with some scenes 22 or 24 vs. 36 has more of an impact in print. But when I tried it as long as I was working with RAWs and scaled the 24 MP file up to 36 MP with light sharpening, the impact in print was negligible. It's not that I could never see a difference, but I had to really be looking for it.

 

[dtaylor]

Negligible or Indistinguishable with most lenses. 18mp aps-c is using the center and see the marginal gain but the 50mp FF would see all the ugly edges.

Most of the lenses I listed do not have ugly edges. With some of those comparisons I cannot say the difference is negligible.

You'd never even get close to getting 50mp of real detail from it w/o the Zeiss Otus or 135mm APO. Sure you'd have a nice big file to downsample but 50---->36mp is already diminishing returns.

You keep claiming this, but what is your evidence? What makes you think the 70-200 f/4L IS or any of Canon's super teles won't show a gain?

 

[dtaylor]

And another person that couldn't photograph the inside of a paper bag

Well, guess I prefer sujets & subjects that actually make the cut for paid for publication.
Should the paper bag ever be a topic I'll defer to your expertise.

Oh really? By all means let us see you work on these publications. We're all dying to see them now that you mentioned them.

OMG dude, seriously? clearly high resolution photography and oversampling discussions are not something you're handling well. Can you just let us adults have a thread to discuss without this nonense? Relax and go enjoy whatever equipment you like.

Yeah...he's not making any friends right now.

 

[NancyP]

My my, we are getting testy during the Photokina countdown....... You might think that we were Canon executives whose bonuses depended on success of their product line.

 

[RLPhoto]

Negligible or Indistinguishable with most lenses. 18mp aps-c is using the center and see the marginal gain but the 50mp FF would see all the ugly edges.

Most of the lenses I listed do not have ugly edges. With some of those comparisons I cannot say the difference is negligible.

You'd never even get close to getting 50mp of real detail from it w/o the Zeiss Otus or 135mm APO. Sure you'd have a nice big file to downsample but 50---->36mp is already diminishing returns.

You keep claiming this, but what is your evidence? What makes you think the 70-200 f/4L IS or any of Canon's super teles won't show a gain?

Visit dustin abbotts review on the APO 135mm and look at other results. DXO is consistent with my uses on the lenses I have and trust their lens results to a degree.

135mm f/2L is no slouch but even that lens can't get more detail at a certain point when the APO zeiss gets much more. Diminishing returns, more expensive glass to get detail in those MP, and certainly without a shadow of a doubt, the lens is the limiting factor for getting what your paying for in a 50mp camera.

So slapping on a 24-70mm on a 50mp camera is not going to look much different than a 36mp camera, which looked pretty nice on a 22mp camera.

 

[psolberg]

I would argue that 22 MP vs. 36 MP is negligible with any lens. You need a 50% or greater gain on each axis before it really becomes evident in print.

having shot both, it isn't. But it depends on what you're doing. If you're web publishing at full HD or approximate, then yes.

I was printing samples to an Epson 3880, some crops scaled to appear as if they were being printed on a larger printer (i.e. up to 36" eq).

I'm open to evidence that with some scenes 22 or 24 vs. 36 has more of an impact in print. But when I tried it as long as I was working with RAWs and scaled the 24 MP file up to 36 MP with light sharpening, the impact in print was negligible. It's not that I could never see a difference, but I had to really be looking for it.

scale up and sharpening isn't going to yield the same benefits than scaling down which is essentially going to oversample the bayer artifacts that plague digital. While what you did was basically create data which asn't there to begin with you also magnified the digital artifacts that degrade IQ. Then there is noise. Noise gets smaller and finer without detail killing algorithms when you downsize but larger and more bloated when you upsize.

Ultimately the technique will dictate your results. Garbage in, garbage out as they say. But if I were to summarize it in one sentence: I rather shoot it at 36-50MP and benefit from downsample all the time than the alternative. I think this why ALL camera makers, including canon will no doubt put 50+MP as the de-facto standard for most of their full frame sensors aimed at landscape and studio.

As for printing, the problem is that you introduce so much variation: inks, paper, humidity, temperature, viewing conditions all can affect your perception of a print output which makes "looks the same to me" comparisons highly irrelevant. However of interest to digital photographers concerned with detail is the RAW file as measure of image quality. How that projects to XYZ printing environment by no means negates gains in the digital files. As I said, I think this is all inevitable this is the way all OEMs will go. Other than storage which is always getting cheaper and buffer and frame-rates limits, there is just no real benefit to lower MP counts at the digital file itself.

There are some interesting entries over at Loyd chambers that discuss improvements in color sampling, color resolution sampling, bayer pattern artifacts, and other topics. Just a few I could find, but many more exist all over the web.

http://diglloyd.com/blog/2012/20120819_6-SonyRX100-sensor-density.html
http://diglloyd.com/blog/2013/20130307_3-oversampling-RX100.html
http://diglloyd.com/blog/2013/20130223_3-lenses-for-high-res-digital.html
http://diglloyd.com/blog/2012/20120209_1-DepthOfField.html

the thing I find confuses a lot of photographers which claim "my L lens won't resolve more anyway" blah blah is that they do not understand the Bayer matrix used in digital. Once you understand how a RAW file is basically composed of mostly green and huge missing parts of blue and red which are "guessed", you can understand why a high resolution bayer sensor can act as an RGB sensor of lower resolution but that yields better quality and noise characteristics. This is REGARDLESS OF LENS. Yes higher quality lens = better results. But oversample even a kit lens will product a techincally superior image to the same lens under a lower resolution bayer. Off course if you shoot with L lenses or any professional lens from any OEM, you're going to get more than just modest gains...exactly why this sony sensor is exciting.

 

[Skatol]

Dear Friends.
This Attachment is the Old News in August 24, 2010 , about Canon New 120 MP sensor = 4 years ago.
BUT, Why, Now 2014, The Best Sensor that on the Market by Canon = only 22.3 MP---- ??, And Top of the Line of Canon = 1Dx only = 18.1 MP..
May Be ( ??) ,Past 10 years, Only 1.23 Billion users of Facebook( Like ME ) have post the Photos on Face book need only 2 MP Cell Phone Camera's photos of Foods that they eat at morning/ Breakfast , to show their friend---NOT 50 MP. Sony Camera, Now.

http://www.dpreview.com/articles/2773376832/canon120mpsensor

Enjoy
Surapon

http://www.theguardian.com/technology/2014/feb/04/facebook-10-years-mark-zuckerberg

Thank you Mr. Surapon for attempting to lighten the discussion a bit. I always appreciate your contributions and the optimism you bring to this forum. BTW, is Biscuitville still in operation down there? They made the best breakfast!

 

[jeffa4444]

I would be very surprised so soon after announcing the A7S and the A7R being so young that Sony would bring out a 50MP CILC / DSLR.
More likely its the 50MP sensor Hasselblad, Phase One and Pentax are using being applied to a RX-1 type camera.

Its not a simple oversample issue either at some point Sony et al will need to move to XQD or CFast 2 cards which are expensive and not widely adopted the Arri Amira being the first camera to use CFast 2. Our PCs will need the newer version of Haswell processors & Quick Sync for 4K video off of such a sensor so the whole thing has a knock on.

 

[jrista]

I would argue that 22 MP vs. 36 MP is negligible with any lens. You need a 50% or greater gain on each axis before it really becomes evident in print.

having shot both, it isn't. But it depends on what you're doing. If you're web publishing at full HD or approximate, then yes.

I was printing samples to an Epson 3880, some crops scaled to appear as if they were being printed on a larger printer (i.e. up to 36" eq).

I'm open to evidence that with some scenes 22 or 24 vs. 36 has more of an impact in print. But when I tried it as long as I was working with RAWs and scaled the 24 MP file up to 36 MP with light sharpening, the impact in print was negligible. It's not that I could never see a difference, but I had to really be looking for it.

scale up and sharpening isn't going to yield the same benefits than scaling down which is essentially going to oversample the bayer artifacts that plague digital. While what you did was basically create data which asn't there to begin with you also magnified the digital artifacts that degrade IQ. Then there is noise. Noise gets smaller and finer without detail killing algorithms when you downsize but larger and more bloated when you upsize.

Ultimately the technique will dictate your results. Garbage in, garbage out as they say. But if I were to summarize it in one sentence: I rather shoot it at 36-50MP and benefit from downsample all the time than the alternative. I think this why ALL camera makers, including canon will no doubt put 50+MP as the de-facto standard for most of their full frame sensors aimed at landscape and studio.

As for printing, the problem is that you introduce so much variation: inks, paper, humidity, temperature, viewing conditions all can affect your perception of a print output which makes "looks the same to me" comparisons highly irrelevant. However of interest to digital photographers concerned with detail is the RAW file as measure of image quality. How that projects to XYZ printing environment by no means negates gains in the digital files. As I said, I think this is all inevitable this is the way all OEMs will go. Other than storage which is always getting cheaper and buffer and frame-rates limits, there is just no real benefit to lower MP counts at the digital file itself.

There are some interesting entries over at Loyd chambers that discuss improvements in color sampling, color resolution sampling, bayer pattern artifacts, and other topics. Just a few I could find, but many more exist all over the web.

http://diglloyd.com/blog/2012/20120819_6-SonyRX100-sensor-density.html
http://diglloyd.com/blog/2013/20130307_3-oversampling-RX100.html
http://diglloyd.com/blog/2013/20130223_3-lenses-for-high-res-digital.html
http://diglloyd.com/blog/2012/20120209_1-DepthOfField.html

the thing I find confuses a lot of photographers which claim "my L lens won't resolve more anyway" blah blah is that they do not understand the Bayer matrix used in digital. Once you understand how a RAW file is basically composed of mostly green and huge missing parts of blue and red which are "guessed", you can understand why a high resolution bayer sensor can act as an RGB sensor of lower resolution but that yields better quality and noise characteristics. This is REGARDLESS OF LENS. Yes higher quality lens = better results. But oversample even a kit lens will product a techincally superior image to the same lens under a lower resolution bayer. Off course if you shoot with L lenses or any professional lens from any OEM, you're going to get more than just modest gains...exactly why this sony sensor is exciting.

Well put. Oversampling is definitely where the industry should go. It allows them to legitimately remove AA filters (as they would no longer be necessary with sufficient oversampling), and it opens up doors for better processing.

For one, we could stop using bayer interpolation. We could do super-pixel interpolation, where each independent set of 2x2 matrix of RGGB pixels is blended together for each output RGB pixel. That would eliminate the softening that arises from interpolating every intersection of pixels, and improves color fidelity.

Another benefit of oversampling is noise frequencies become significantly smaller than the smallest detail frequency. That makes NR far more effective...you could apply heavier NR (preferably some kind of wavelet or generalized variation NR) that could completely obliterate noise entirely, while having a minimal to insignificant impact on detail itself. When you downsample, your now downsampling an image that is starting out with pretty much no noise...then averaging pixels together to reduce any fractional remnant of noise even further.

There are so many benefits to oversampling.

 

[dtaylor]

DXO is consistent with my uses on the lenses I have and trust their lens results to a degree.

The page you've linked to at DxO does not support your claim.

So slapping on a 24-70mm on a 50mp camera is not going to look much different than a 36mp camera, which looked pretty nice on a 22mp camera.

But there is going to be an observable gain.

 

[dtaylor]

scale up and sharpening isn't going to yield the same benefits than scaling down which is essentially going to oversample the bayer artifacts that plague digital.

Scaling 36 MP down to 24 MP and then printing will pretty much throw away whatever advantage the 36 MP file had.

While what you did was basically create data which asn't there to begin with you also magnified the digital artifacts that degrade IQ.

So in theory this should have been a worse situation for the 24 MP files, yet it resulted in nearly identical prints. I could try printing them without first scaling the 24 MP file, but I don't expect a difference. The reason I scaled in the first place was to more accurately judge at pixel peeping size on my monitor. You have to do this or psychologically the "bigger" image will impress you more.

Ultimately the technique will dictate your results. Garbage in, garbage out as they say. But if I were to summarize it in one sentence: I rather shoot it at 36-50MP and benefit from downsample all the time than the alternative. I think this why ALL camera makers, including canon will no doubt put 50+MP as the de-facto standard for most of their full frame sensors aimed at landscape and studio.

I think the future holds >50 MP sensors as well and for similar reasons. I do not deny that oversampling can yield benefits.

As for printing, the problem is that you introduce so much variation: inks, paper, humidity, temperature, viewing conditions all can affect your perception of a print output which makes "looks the same to me" comparisons highly irrelevant.

A 3880 with Hot Press Bright is more than capable of resolving the finest detail if we're talking about 16x24" prints or crops scaled as if they were being printed on Epson printers capable of 24x36". If I have to pixel peep to see a difference, but can't see the difference at smaller scales on screen or especially with that printer/paper combo, then the difference means nothing to me.

However of interest to digital photographers concerned with detail is the RAW file as measure of image quality.

Pixel peeping at 100% showed very little difference in the test scenes I reviewed. I could pick out 2 or 3 small areas where the A7R file resolved some fine texture or detail better, but >95% of the scene was identical.

 

[jrista]

scale up and sharpening isn't going to yield the same benefits than scaling down which is essentially going to oversample the bayer artifacts that plague digital.

Scaling 36 MP down to 24 MP and then printing will pretty much throw away whatever advantage the 36 MP file had.

Sorry, that's totally false. Downsampling 36mp to 24mp produces significant gains in the quality of the smaller image, and if your not printing it at a size that requires more than 24mp, that's not a loss. That IS one of the advantages of having a 36mp image.

Furthermore, if you need to upsample, starting with more data means the upsampling process is more accurate, as it has to fabricate LESS artificial data than s 24mp or 20mp or 18mp file.

Ultimately the technique will dictate your results. Garbage in, garbage out as they say. But if I were to summarize it in one sentence: I rather shoot it at 36-50MP and benefit from downsample all the time than the alternative. I think this why ALL camera makers, including canon will no doubt put 50+MP as the de-facto standard for most of their full frame sensors aimed at landscape and studio.

I think the future holds >50 MP sensors as well and for similar reasons. I do not deny that oversampling can yield benefits.

I doubt it will stop at 50mp. I think we'll hit 100mp in a camera that actually hits store (or warehouse) shelves at some point. I think we will eventually also see 150mp cameras...at some point the high end BSI technology that drives <2µm pixels in small sensors will be refined and perfected enough that the substrate fragility is no longer a problem for larger sensor areas...then...why the hell not?!? For those not concerned with disk space, the higher megapixel count and oversampling would offer significant benefits. For those who are concerned with disk space, there is no reason super-pixel demosaicing to produce a smaller, but still high bit depth data file couldn't be used in-camera to reduce megapixel count considerably, saving space and transfer time.

As for printing, the problem is that you introduce so much variation: inks, paper, humidity, temperature, viewing conditions all can affect your perception of a print output which makes "looks the same to me" comparisons highly irrelevant.

A 3880 with Hot Press Bright is more than capable of resolving the finest detail if we're talking about 16x24" prints or crops scaled as if they were being printed on Epson printers capable of 24x36". If I have to pixel peep to see a difference, but can't see the difference at smaller scales on screen or especially with that printer/paper combo, then the difference means nothing to me.

Hang out on a print forum sometime. For all the bickering we do over pixels, they do 10 times more over paper quality, ink quality, dMax, L*, metamerism, bronzing, ink emulsions, etc. It matters to the people at the lab who do all the printing for...millions of people who order prints from labs. If it wasn't for the IQ nutbags creating and running print labs, inkjet technology probably wouldn't have come as far as it has. Demand for more and better keeps markets moving and improving.

However of interest to digital photographers concerned with detail is the RAW file as measure of image quality.

Pixel peeping at 100% showed very little difference in the test scenes I reviewed. I could pick out 2 or 3 small areas where the A7R file resolved some fine texture or detail better, but >95% of the scene was identical.

You yourself know it isn't just about looking at two images strait out of camera. The editing latitude differences matter as well, the response to things like NR or white balance or exposure changes that better data with richer information that gives us the ability to reduce noise more effectively with less effort, that all matters. Maybe not to everyone, maybe not even the majority (at the moment), but to a lot of people...to an increasing number of people, IMO.

 

[RLPhoto]

DXO is consistent with my uses on the lenses I have and trust their lens results to a degree.

The page you've linked to at DxO does not support your claim.

So slapping on a 24-70mm on a 50mp camera is not going to look much different than a 36mp camera, which looked pretty nice on a 22mp camera.

But there is going to be an observable gain.

The DXO info perfectly supports my claim. Jumping to a 50mp sensor, your never going to get close to that resolution with most of your lenses.

 

[jrista]

DXO is consistent with my uses on the lenses I have and trust their lens results to a degree.

The page you've linked to at DxO does not support your claim.

So slapping on a 24-70mm on a 50mp camera is not going to look much different than a 36mp camera, which looked pretty nice on a 22mp camera.

But there is going to be an observable gain.

The DXO info perfectly supports my claim. Jumping to a 50mp sensor, your never going to get close to that resolution with most of your lenses.

You might be trusting DXO's numbers too much. Their primary results are usually done at very aberration-limited apertures, and rarely seem to jive with the earliest diffraction-limited apertures as determined by other testers. Plus, lenses are entirely non-linear in their performance curve. The same lens could perform poorly at f/2.8, superbly well at f/4, and ok at f/8. If you have the option of using f/4, you could realize significant gains by moving from 36mp to 50mp. If you have to use f/8 for DOF, you might not realize as much of a gain (although you would be getting more into the territory of oversampling...and we've been discussing the benefits of that). At f/2.8 your unlikely to see any real improvement in resolution...however, fast apertures are often more about the potential aesthetics that certain optical aberrations bring to the table anyway...so whether it resolves more or not isn't the point. (And, your still going to be sampling the lens better with the 50mp sensor...so your never losing anything, all that changes is how much your gaining...but it is always a gain.)

Assuming were using a quality lens at f/4 that is diffraction limited. That's 173lp/mm. A 50mp sensor is probably around 120lp/mm. A 36mp sensor is 102lp/mm. The 36mp camera with that f/4 lens is going to produce an output resolution (the spatial resolution of your RAW images...not the pixel resolution) of ~88lp/mm. The 50mp camera with that same lens is going to produce an output resolution of ~98.5lp/mm. That is a difference of 12%. That is not insignificant...it's quite significant. It's well beyond the margin of error.

A 70mp sensor would resolve ~109lp/mm. At this point, this ultra high resolution sensor is resolving more than the 36mp sensor ever could. We are still well below the resolving power of the lens, which means we are still undersampling...which means there are still gains to be made.

The next step would be 100mp. At 100mp the size of our pixels is about the size of a diffraction spot at f/4. Were now sampling at the same rate as the lens. We are not oversampling yet...were resolving 122lp/mm...which is still below the resolving limit of the lens. We could keep going...and still realize gains. At 150mp...200mp...maybe around there we run into some limitations with physics, making the cost of pushing pixel sizes smaller too high to be cost effective for the end user. At aberration limited and smaller diffraction limited apertures, we are oversampling now...we'd be oversampling anything smaller than f/4 by 100mp. A lens like the Otus, however, would still be capable of resolving a diffraction spot considerably smaller than a pixel at apertures wider than f/4...and then were talking bout 240 to 400 lp/mm up through around f/1.4.

There is always room for more resolving power. It may not always be useful to everyone...if you live and die at f/8, you might well already be oversampling to a useful degree with pixels around 4µm. If you are an Otus fan, which is probably capable of resolving 200-300lp/mm at wider apertures, then a 50mp sensor is nothing...you could probably use a 150mp sensor and still not come close to oversampling the lens at it's best.

 

[dtaylor]

Sorry, that's totally false.

Go try it yourself. There are A7 / A7R test scene samples at multiple sites.

Furthermore, if you need to upsample, starting with more data means the upsampling process is more accurate, as it has to fabricate LESS artificial data than s 24mp or 20mp or 18mp file.

That's true in so far as there's more real data (lpmm at a given MTF) to begin with.

Hang out on a print forum sometime. For all the bickering we do over pixels, they do 10 times more over paper quality, ink quality, dMax, L*, metamerism, bronzing, ink emulsions, etc. It matters to the people at the lab who do all the printing for...millions of people who order prints from labs.

Any of the current Epson Professional series printers/ink sets on a paper like Hot Press Bright...you're looking at the very top end of what can be laid down on paper today. I won't order laser photo paper prints (Frontier or Noritsu) when I can print or order Epson.

You yourself know it isn't just about looking at two images strait out of camera. The editing latitude differences matter as well, the response to things like NR or white balance or exposure changes that better data with richer information that gives us the ability to reduce noise more effectively with less effort, that all matters. Maybe not to everyone, maybe not even the majority (at the moment), but to a lot of people...to an increasing number of people, IMO.

There's not a significant difference between an A7 and an A7R in these respects.

 

[dtaylor]

The DXO info perfectly supports my claim. Jumping to a 50mp sensor, your never going to get close to that resolution with most of your lenses.

You didn't link to anything that supports this claim. For however significant you consider the jump to 50 MP in the first place, that gain is not going to be muted or blocked by better lenses. Glass is not (yet) the limiting factor, at least not at the upper end.

 

[RLPhoto]

The DXO info perfectly supports my claim. Jumping to a 50mp sensor, your never going to get close to that resolution with most of your lenses.

You didn't link to anything that supports this claim. For however significant you consider the jump to 50 MP in the first place, that gain is not going to be muted or blocked by better lenses. Glass is not (yet) the limiting factor, at least not at the upper end.

I did link it previously and only the zeiss APO 135mm was able to get the maximum 36mp out of the camera. A lowly 24-70mm wasn't even close to getting the details out of the d800. You should check the link.

To really get 50mp worth of detail, don't expect anything less than the most expensive glass to achieve it. Sure, I understand there will be gains but in my original post here, I stated most lenses will never get close to resolving the full 50mp in that sensor. Just like most lenses won't get even close to resolving the majority of the 36mp in the current d800.