As I guess most of you find out by now the previous rumor about the A91 and A9 cameras was a fool days joke 🙂 But I have a bad and good news for you:
Bad: The rumor is obviously false. There will be no A91 and A9 on May 1.
Good: Foveon sensor, 50+ megapixel sensor, Super fast AF. All the info I shared is based on rumors I really got. So there is a big hope Sony will make that fool camera specs…for real!
–
And now I am at work to get and post some “real” rumors 🙂
0 Responses
”
Good: Foveon sensor, 50+ megapixel sensor, Super fast AF.”
Time to look at nikonrumors. 🙂
just want to know will the next a-mount be a mirrorless(without slt)
me 2
I’m not sure the tradeoff is worth it – slower af, inevitable compatibility issues with 3rd party glass vs 1/2 stop light boost. If they could deliver something mirrorless with current gen slt af performance, but without any compatibility issues, then I’d be all over it. My guess is we’ll get an improved slt mirror this time round though, which is fine for me.
The a6000 has fast af although and is mirrorless.
Not with legacy lenses.
There is no substitution for speed. The new SLT cameras are going to be much faster than the a6000.
I hope the medium format rumor is true
This is still a joke because you still use the word Foveon!
I heard Dongbu HiTek is about to sell their Foveon factory.
Sony did anounce they expect sensor sales to exceed production capacity by 2015..if the price is right who knows?
It’s just a factory and it will still be working after being sold. Foveon has the patents.
Sigma has the Foveon patents, I mean.
multi layer sensors are crap for hi ISO
With time, that will change. It’s already improved on the dpq series, at least according to Sigma.
Dear friend! Teach yourself a logic, please. I would not even mention with multilayer sensors we can reduce light loss (one pixel consumes all the visible light, not the certain part of the visible spectre) and thus can have better high ISO.
There is a paper by Eric Fossum which suggests 2-layer sensors are the way to go.
If the pixels indeed were capturing all the light that would ne true. However, if we use the Foveon-tehcnology, at lest half of the silicon of the pixel (depthwise) is “dead-zone”, where photons are lost – without this the photodiodes would not work. Then you have much more metal wiring than conventional sensors have which loses even more light as photons hitting them are lost.
Now, if you shoot in color, the color conversion of Foveon-technology data creates massive amounts of noise.
Also, correlated double sampling is not possible with the bottom two layers, thus the read noise will be an order of magnitude higher than what the competition has.
Also you just might have a look at the actual data or even images from Foveon sensors.
Now, maybe in the future we will have organic film which would eliminate much of the issues, though CDS would still likely be an issue, but with thrice more data the read noise would not be as much an issue. But that technology is nowhere near maturity – maybe 10 years.
Sony should use Color Splitter Tech that Panasonic has developed instead of multilayer : http://www.engadget.com/2013/03/31/video-explains-how-panasonics-color-splitter-sensor-works/
all the benefits of Bayer – with 70% more light !!!
Picture restoration after microprisms is extremely complicated process. In general each restoration costs resolving of the system of linear equations with the dimensions of megapixel count, i.e. 18 million, 16 million, 24 million. In this particular case the system has stipped view matrix that significantly reduces computational complexity, and if 18 million vs 18 million system resolving is practically impossible, the special view makes it is observable: it is «only» matter of minutes. Not fast enough, isn’t it? 😉
Cameras today have very heavy algorithms that does a thousand things to the image from its read in RAW pixels to the output JPEG – including noise reduction which in the past was a heavy task for a normal CPU.
NoiseNinja could easily spend 1 minute on my old Pentium 4 machine – but A99 with BIONZ does practically the same in miliseconds.
Also, the operation cost from reconstructing images from an encoded format like the colorsplitted image, is roughly the same cost as BAYER to RGB = no problem
Probably not stupid, and clearly a better grasp of math and electronics than you.
Please. Different algorithms of NR are obviously utilized in this case. OOC one is based on frequency thresholding, which is fast but not very accurate and smear details. Ninja is on some advancement of non local means, which is slow, but much more efficient.
And, second, you need to show LV. At least tens of frames per second.
Demosaic is much easier procedure than reconstruction of splitted pixels image: O(N) vs O(≥N^2.8), where N is Mp count.
According to the article
http://blog.gsmarena.com/video-details-panasonics-color-splitter-tech-that-promises-big-improvements-to-digital-camera-sensors/
The RGB values are easily calculated from the matrix (give or take some minor adjustments) – so i don’t see what the fuss is all about.
No, of course that is not that simple. First, you have neighbour pixels’ contribution into the value collected of each photodiode and you need to estimate what piece of value refer to each pixel. This will lead to equations resolving.
why do you say O(N^2.8) and N is Mp count whereas splitters work only in scope of their row. It’s not going to be much more difficult to demosaic than bayer.
Nobody would ever need exact solutions of system of linear equations, it more needs to be stable than exact. In practice, few iterations of convolution will be enough for liveview.
> it more needs to be stable than exact.
Indeed. It can be done stable with special edge prisms, that should not spread a light beyond the sensor borders.
What are you thinking of? Do engineers need to invent special prisms just because someone else wants do use unstable demosaic method?
There will be losses anyway, what about noise, dead, hot pixels, dust?
It doesn’t have benefits of bayer.
First, its demosaic will need SUBTRACTIONS (like CMYG does need), so you’re getting less luminance noise at more chroma noise.
Second, it will suffer from colorshift and color response issues.
Third, color splitters will interfere with microprisms so um.
you are correct, it was wrong to say “all the benfits of Bayer” – since bayer isn’t actually good, due to multiple issues.
what i meant was, that Panasonics tech would work AS GOOD AS Bayer, only 70% better in regards to capturing light.
Foveon would still be the better tech than everything else.
Apologies if this question sounds so basic that it borders on stupid….
If I get this right, a 50MP Foveon sensor yields a megapixel count of 50 divided by 3?! In other words: About 16-17MP final output in terms of image size?!
I guess so.
A 54MP Foveon sensor actually produces an 18MP photo. The great thing about this, is that each pixel gets RGB information. The bad thing about it, is that only the Blue layer gets the entire light coming from the lens. The Green layer gets less light, and the Red layer gets the least light.
On the other hand, a traditional Bayer sensor, has GGRB configuration, meaning 2 pixels are sensitive to Green, 1 pixel is sensitive to Red and 1 pixel is sensitive to Blue (this pattern continues across the entire sensor). The good: each pixel gets the entire light coming from the lens. The downside is that each pixel gets only one primary color, and not all three.
The industry has obviously chosen the latter.
A camera based on a Bayer-sensor always has to de-bayer those pixels, and this is where manufacturers can use different de-bayer algorithms, hence you get different “JPG engines”.
The yellow pigment absorbs ONLY blue and is virtually 100% pure
Magenta absorbs green and is about 94-97% pure.
Cyan ia less pur..only about 86% pure but is in the last position.
There is virtually no difference betwwen the light passing through mutiple layers in the stacked sensor and the individual filters in bayer.
The impurities are compensated for in a process called color correction which removes the colors that dont belong.
yep, but the quality is about twice that of bayer at same res, and it can resolve detail equaly in all colours, unlike bayer.
with the no AA filter and demoisacing, there very very sharp, whole other level of sharpness 🙂
sigma dp series is pretty amazing 🙂
Resolution tests give about 10% advantage. Any more is due to the anti-alias filter.
Without the AA-filter there is aliasing – what you perceive as sharpness is actually flawed image, a lego-brick representation of analog nature. For some subjects and people it may look goog, for some not so.
Also the resolution of the layers vary significantly in spite of the same pixel count – the bottom layer is by far the least sharp, suffering the most from cross-talk. This is especially issue when you leave the center of the sensor area. The top layer is the sharpest.
Other qualities of the Foveon sensor as quite apalling – there are reasons why this kind of technology is very very niche.
You may like the dp-series lots, but truthfully, any E-lens camera and modern m4/3 camera is much more splendid all-around camera even if we talk strictly about image quality.
I think black and white test charts are more favourable to bayer, on real subjects, the dp series really shines, really pulls out details in a honest way, sort of like hdr, or using lots of clarity, but in a way that feels right.
as to the lego brick thing, yeah thats kinda true, but a bayer can mush detail, and is made up of the same pixels anyway, just smoother. and if i want that i can just doublesize the sigmas files, thatll give it a bayer look to it at 100% (200% effectively)
I prefer the sigma dp black and white to the leica monochrome too, having tried one out briefly.
Its 54mp 3 layers of 18mp.
The MF format sensor better be for real. I’m tired of these immature April fools hoaxes. * Yawn *
The most important information, do these 3 items combine together?
I just want one A-mount Mirrorless FF (Big Body)
It will be the real A900/A850 successor~
haha good one! that’s why I didn’t even bother to comment on that story 😉 I take everything with a grain of salt near April fools! lol
I heard they’re bringing back the A6x models. Will announce A69 soon.
I would not be surprised to see a return to dichroic mirrors, a low light loss way of spliting an image into three separate paths for each primary color and a dedicated sensor for each path. This is how the first color TV cameras worked back in 1939!
Not sure if they use prisms or mirrors but you can certainly buy 3-sensor cameras today if you want. Look at the high end of the Sony video line for examples.
Interesting to know this old technology is still in use in high end cameras. Thanks.
I got interested, but I don’t see any on Sony’s website. Anyways, I believe they’d be prism based. Prism based light separation is 2x-3x more compact than a dichroic mirror setup.
Not too many left in the line-up. The Sony HDW-F900R is still for sale. Price is around $80,000 no viewfinder or lens.
This would mean physically huge lenses, large (thick) cameras and inaccurate colors when the spectrum is nor reletively wide and even. You’d win in sensitivity as well as tonality though.
Will never happen.
It looks like you’re correct, as evidenced by the huge cameras Fred4d pointed out.
You shouldn’t say Foveon sensor when you mean layered sensor. Foveon (nowadays a subsidiary of Sigma) isn’t the only company having patented such a design. Both Sony and Canon are working on layered sensors too.
The first priority of Sony is to make more lenses ASAP. She may also encourage the third party manufacturers to make more lenses for our choice. That’s meaning less to lead the pixel war because it is just pre-consuming the path of upgrading for the second generation product. Sony should admit that Fujifilm has beaten Sony on noise reduction, Olympus leads the technology on 5 axis stablizer and a long way to get close to the AF speed of DSLR. I own A7R, but I have to admit excect pixel and FF, that’s nothing surprise me and love it deeply since I have tried all other better cameras before. The only reason I own A7R is because it is able to it fit in all the lenses in manual focus. However, it is not modern technology.
Why more lenses? They are big, bulky, ugly and expensive.
It’s way better to have shiny new cameras every three months or so.
:o)
If you have enough MP’s then all you need is the 55mm prime and you can crop the image to the equiv. of 550mm 🙂
Actually if you have enough megapixels, all you need is a 21mm prime (Otus class) and you can crop to any narrower field of view. A 55mm can’t get you wide angle, but wide angle can get you 55mm FOV. (without the bokeh)
The whole stochastic pixel distribution approach seems to be getting lost in the battle between bayer and layered.
What battle? Sigma sells maybe 10.000 cameras a year. Conventional sensors sell maybe 1 billion a year. Sigma and Foveon and the whole cocept of using silicons inherit properties to separate the signal energies weakly is extremely niche with no viable future.
I really want a 24-70 2.8 OSS FE lens. I’m willing to pay 2000 for it. 🙂
If this lens come out, I think I will stick to e-mount. Right now I’m planning to buy the LAEA4 adapter and adapt the sony zeiss a mount lens. The ones I’m looking at are 24-70 2.8, sony zeiss 50 1.4 and 105. I got rid of all my Canon stuff when I bought the A7. I’ve been using nothing but the kit lens and to be quite honest, I’m really really happy with the kit lens. I’ve shot with the 5d3 and canon 24-70 but it doesn’t look as nice as what I was able to shoot with the a7 kit lens combo. 🙂
Some critical thinking before releasing SR0 sumors as SRx… Foveon sensor is extremely sensitive to the chief ray angle – impossible to use with close exit pupil – even SLR normal lens exit pupil is problematic. The color separation is extremely weak and leads to massive increase in noise for color photography as well as metameric failure issues. Correlated double sampling is impossible to achieve to the bottom two layers which increases read noise by an order of magnitude.
So to increase resolution by 10% or so, you’d be sacrificing dynamic range, low light performance, color accuracy and increase data processing need as well as amount of it as well as increase the manufacturing costs significantly and make the old lenses for E-mount unusabe.
Let’s just agree you doubled down on april fools rumor…