abera wrote:
Scottelly wrote:
abera wrote:
dellaaa wrote:
Reading a recent post that included a 60M pixel Bayer sensor shot, the the discussion turned into a debate as to whether the FFF would ever arrive.
Looking at the Bayer photo what do you believe the FFF sensor would have to do better than the current generation of Bayer sensors? What are the design criteria the FFF design team have?
I don't know but given the physics of the design and the past Foveons, low light performance would most likely not be on par with the Bayer, agreed?
Foveon has three photodiodes per pixel, none with correlated double sampling. This means very high read noise which means high noise floor which mean poor performance in areas with little light. This means poor low light performance and relatively low dynamic range.
If that's the case, then why did early Foveon sensors have so much more dynamic range than the Bayer pattern CFA sensors of their time?
Did they? Can I see some measurements proving it? I have never seen any.
The issue with read noise can not really be solved
Why not?
Two reasons:
Three photodiodes per pixel instead of one. This alone means sqrt(3*n^2) of noise per pixel compared to single photodiode.
The other is practical impossibility of CDS.
- there will naturally be more photodiodes, but also CDS is quite impossible to achieve with this kind of design.
What is CDS?
Correlated double sampling. It's used to eliminte the main pixel read noise element, the reset noise. With the Foveon vertical pixel structure it's hard to see this happening.
Some fancy non-destructive multiple sampling might be possible but it would not only slow things down even more, but also not be as effective.
From QE point of view the advantage of Foveon is not having colour filter array blocking light, but on the other hand there is plenty of dead space between the photodiodes. I'd guess one might be able to achieve slightly larger QE with Foveon which would help somewhat in low light situations, but not with the noise floor itself.
Historically, Sigmas have required extensive image processing both while taking the shot and afterwords (SPP). Will the FFF shoot at the frame rates of the current Bayers? Again, given the history, this seems doubtful.
Foveon's problem is the very poor color separation.
Huh? I thought it was supposed to be noise that was Foveon's "problem."
This was in the context of the paragraph.
Read noise isn't really a big problem for most photography as the maximum dynamic range is rarely needed (especially for JPG shooters it's quite irrelevant). Though for low exposure photography (e.g. hand held low light) this is an issue.
Color separation issue is on the other hand always there.
Foveon separates colors by taking advantage of the fact that different wavelengths of photons tend to be absorbed at different depth in silicon - the problem is that this is not a deterministic process, but a probabilistic one making the color separation very weak.
Also the photon absorption is quite a bit top heavy - as you know Sigma made the two bottom layer photodiodes four times the size of the top layer (areawise, depthwise they are also much larger). There is a reason for it - the bottom layers are simply relatively photon starved. Additional benefit is reduction in total read noise. It also indicates the difficulty of increasing the resolution of the bottom layers from image quality point of view - I think this is a good compromise even though it's a deviation from the original design, increases SNR somewhat with neglible resoltution loss, though maybe opens the door slightly to false color artifacts.
In principle this could be helped by placing narrow band color filters on top of the sensor, but then some wavelengths would be totally lost and could cause some interesting artifacts.
To get decent color out of Foveon the processing indeed will be rather strong - this increases noise for color photography.
Where do you get this idea?
From the weak color separation.
Page 177 of this . Note the spectral sensitivity chart and the color matrices.
So with low light and speed out of the equation, whats left, detail and color rendition.
The main advantage is reduction in false color artifacts (e.g. moiré). This advantage goes away once diffraction (and lens deficiencies) does anti-aliasing, but it's still a long way from today.
Detail situtation has many variables, but regardless of them lots and lots of pixels would be needed in this hypotetical FFF to compete.
Color accuracy is vastly better on conventional sensors. That's why the CFA is there for.
Wrong. The Quattro generation was one of the most color accurate sensors of its time.
Wrong. From the above document:
One does not get more accurate colors from exremely weak color separation. One gets less accurate colors with more noise.
Technology has moved on, but for the money you probabky coukdn't get a camera with a more color accurate sensor than a Quattro.
Likely all conventional Bayer CFA cameras have more accurate colors from hardware point of view.
Well if that is true, then you're saying Sigma knows how to coax better color out of sensors than Nikon. I find that hard to believe, but I guess it's possible. There have been several fp L users here who have multiple systems, and they prefer the colors they get from their Sigma cameras.
Other small companies have been known for their color too, such as Leica, Hasselblad, and Phase One.
Foveon type multi layer system might work better in the future with different materials, maybe organic materials or perovskites. Plenty of perovskite developement is being done due to their potential for solar cells and at least one multilayer demonstration does exists (with hideous image quality). Maybe 15 years from now?
As for detail, how would the specs of the proposed FFF stack up against the 60M plus Bayer sensors? Would the advantage in detail make up for the shortcoming?
No. You need likely something like 70% of the pixel count to match conventional sensors, maybe more nowdays as AI-based demosaicing is on its way. This also diminishes the false color advantage.
Anyhow, as even 10MP is enough for very large prints...
The remaining design criteria, is color. Would Sigma's color rendition be enough for people to buy it?
Some like it's different colors and I'm sure many fans would buy because of them, but as the colors can be very problematic and lack in accuracy and because the conventional cameras are not exactly lacking in color accuracy, it's hard to see how the general public would get interested in this.
I don't know and would like rational, opinions as just what will the FFF bring to the table, and why any non L mount people in 2023 would buy it when 100M pixel Bayers are on the horizon that take great low light picture at high frame rates.
At this point, fanboys aside, the fact that we are still discussing this seems a bit absurd to me, so why did I start this thread lol?.
Foveon is very interesting concept and I'm happy it was brought to life. But it feels more like a hobby project, a niche project for those who desire something different. There aren't really any advantages over conventional sensors
Huh? You already gave one in your own reply here above . . . "The main advantage is reduction in false color artifacts (e.g. moiré)."
I said "really any advantages" - meaning that there aren't much relevant advantages. But I can agree that I should have used more careful wording.
As long the the sampling rates of conventional sensors are low enough for diffraction (and lens flaws) to not be enough to eliminate false color artifacts, this can be a small advantage.
Of course it could esily be corrected by using strong enough anti aliasing filters, but since pixel peeping is part of camera reviews and those sell cameras, AA-filters are now largely gone or very weak (and often unidirectional as well.
while the drawbacks may cause issues.
I hope a FF Foveon does see the light of day, but not only I'm not holding my breath, I'd expect the product to be a very low volume very high price item.
Of course the first FFF will be low volume and high priced, but for all we know Sigma will make a future generation in ten times the volume
How? In order for Sigma FFF be a large scale products it would need to have tangible advantages without tangible disadvantages. This is not the case. Also it would need to offer similar marketing (including indirect marketing like camera clubs, peer pressure etc.) and support networks, camera shop visibility and so on. There is no reason to believen any of this will happen.
There have been many products put on the market, which dominated a niche in their industry without a large support network or a significant marketing presence. The product just has to work well. Today, with the Internet, that is even more likely, because of how easy it is for people to find the information.
, allowibg them to price it very low, so it sells quite well.
So it will sell well because it's price will go down because it sells well? Something circular in that argument.
No. It will sell well, because its price will be set low by the manufacturer from the beginning, because the manufacturer doesn't need to sell it for such high prices, because it costs less to make, since it is being manufactured in higher volume.
You seem like a pretty well informed person. It's difficult for me to believe you don't know how this works. Your comment brings into question your motives, and makes me wonder if you just have an axe to grind. Are you just trolling here?
Do you know Sigma's history in the camera world? Do you realise they've been selling cameras for about 20 years? Do you know Foveon sensors have been improved time and time again, and Sigma cameras have continued to offer some of the best quality digital imaging, against the massive industry for all those years? I call that amazing, and I prefer the images from the SD Quattro H to everything else, except a 100 MP (or more) medium format camera, as do several other photographers here, despite the dynamic range disadvantage, and despite how slowly the Sigma cameras operate. That alone should tell you something. I have used Canon, Nikon, and Sony cameras (several from each brand), yet I prefer to shoot with Sigma cameras. That's not because I'm stupid, ignorant, or a masochist, but because my Sigma cameras produce better images.
First it would need to exists - if it will ever exist, the long depelopement doesn't really indicate that the mythological "next generation FFF" following it would be developed fast or that it would have enough tangible advantages when the competition will at that point have 200MP or more and be cheaper than it's predecessor.
When even medium format seems to have stopped moving forward in resolution, what makes you think that will ever happen?
Panasonic just made two new cameras, and were either of them high-resolution, with a sensor that offered more detail capture than their first high-resolution full-frame camera? No. How about what Nikon, Sont, and Canon have put out recently? No.
If 200 MP sensors are coming, why are there no full-frame cameras from Canon or Nikon with sensors over 45 MP, even though Canon made a 50 MP camera years ago, and Nikon made their first 45 MP camera (the D850), which was a DSLR, years and years ago?
I'm not suggesting Sigma will fix all the problems with their in-house sensors, but thise sensors have improved in the past, and I have no doubt they will improve in the future too.
Sure. But the last new Sigma sensor came out when? Do you think Sigma has the resources to compete with the rest of the world - Sony and Samsung and others pour billions after billions into R&D getting constantly not only more refined products, but also all kinds of innovations.
I think there will always be a market for good quality, unique products, which offer something different.
Of course - I don't think I said anything contrary. I think I said something quite similar actually
But if you believe that Foveon sensor is some kind of game changer, I'm afraid you're significantly in error.
Leica keeps making monochrome cameras. That is a niche product, but obviously a successful one, or they wouldn't keep doing it.
Sure - though Leica has extreme brand value which gives them possibility of making that kind of "niche of niche" product. It's also dirt cheap for them as sensor without color filter array isn't any more expensive than one with it - likely slightly cheaper even in relatively low volumes and the revamped image processing is a job of a day or so for a software engineer. (They may have done more, but that's the general idea.)
Anyhow, Pentax is another intereting brand - it seems to become the one last DSLR manufacturer providing a DSLR "difference" in mirrorless world. It likely provides modest but solid income to Ricoh and will likely remain a good little niche.
I believe Sigma will continue to improve on the Foveon sensor, making more "niche" products with them in the future, and differentiating themselves from the competition by doing so.
Sure it's entirely possible and even likely. And i hope so. But it'll never be more than a niche player with this technology and an additional problem is that the relative advantage(s) of this technology get smaller and smaller every year. Also I doubt it'll ever try to go head to head with the big boys as the investments would be significant and the risks large. The competition is stiff and the markets aren't as good as they used to be before mobile phone camera revolution.
Sigma is a leading lens manufacturer. To me the cameras are a love child, a hobby - it's (AFAIK) a privata corporation, so they can afford to have such proyects. I do hope Foveon keeps on being developed and produced just for the sake of uniqueness. It makes the world richer. Though, in time it'll lose the false color advantage fully and then it's just game over. This is not that far away, thus investing heavily in dead-end technology is not likely, even if it is a hobby.
FWIW, here's a link to an interesting image about the vertical structure from Chipworks. The comments by Eric Fossum may be of interest. (EF is the inventor of for example 4T pixel structure commonly used in sensors.)
This might help you understand Sigma cameras:
https://medium.com/ice-cream-geometry/an-ode-to-the-sigma-merrill-506dd0864169
