I've only had one Tokina lens, the 11-16 f/2.8 for crop, but it was excellent and produced many of my favorite images. At the moment I have two Tamron lenses, the 70-200 f/2.8 VC G2 and the 45 f/1.8 VC. Love them both, and consider them 'L quality' in all respects. When I finally add a FF macro I'm pretty sure it will be Tamron's.
SIGMA releases an official update about their Foveon X3 sensor project
- Thread starter Canon Rumors Guy
- Start date
I had a Tamron 90mm macro (circa ‘97) that was great optically. The AF was very slow didn’t matter too terribly much for macro. I kept it until a couple of years ago and sold it for a song due to non-use. I sort of wish I had kept it considering what I got for it.
I think it's obvious and evident (what you stated) no need for spreadsheets. Their glass has truly made quantum leaps in terms of optics and build. The stumbled a bit with the launch of the Art series but after they built the 24-35 (completely underrated zoom which could easily replace 3 primes) they started to get it right and with the 85 and 135, they nailed it. I keep waiting for a Foveon travel body.....
As a landscape photographer the Foveon sensor has always interested me. The increased noise at >iso 400ish (if still true) isn’t such a concern for the photography I do. I’m interested to see what they come out with.
Having said that, beginning with the 5d4 I have found myself venturing into the iso 1000 range under certain conditions (as a tool to hit a target shutter speed on moving water). While it may just have been my growing knowledge of post processing I don’t think that would have been possible with the 5d2 I used previously. As I recall 400 was about the max I felt like I could get away with on that body. Limiting factor overall noise but particularly the amount and type of shado noise. The 5d4 was a dramatic improvement in this area.
Still, the potential color/noise performance at lower speeds with foveon tech is interesting to me.
I believe most of Sigma's profits would come from sale of lenses to third party customers rather than selling their own cameras (I wonder how their L-mount cameras would sell). Therefore, I think they would rather prioritize producing Canon RF/Nikon Z lenses.
I tested sigma 35, 50 and 85. The 35 was the best focusing and produced excellent pictures. The 50 was nice, focus issues. The 85... Very bad focus.
I have the ef 35ii 1.4 and rf 85 1.2. Much better focus and great colors
I have the ef 35ii 1.4 and rf 85 1.2. Much better focus and great colors
A 20mp Foveon sensor is closer to a 30mp Bayer sensor. Let’s not overdo it.
The Fiveon sensor isn’t some technological miracle. It has its flaws which have been reported upon. I was given the very first camera that had a Foveon sensor, way back. It was sharper (but not exactly). But it has many flaws, some of which can’t be fixed. The biggest is that colors are “poisoned” by light leaking through the layers. The sensor depends on silicon absorbing different colors of light the deeper they penetrate. But it’s not perfect, and so the layers have amounts of the wrong color light mixed in with the expected color. There is no known way to eliminate this problem. As a result colors are less than pure. Most noticeably in flesh tones, where we’re most sensitive to this.
as far as sharpness is concerned, the Foveon is sharper in red and blue, but less sharp in green.
as I said, it’s far from perfect.
as far as sharpness is concerned, the Foveon is sharper in red and blue, but less sharp in green.
as I said, it’s far from perfect.
No, a 20mp Foveon sensor has the resolution of a 20mp Bayer array sensor with no AA filter, which looks like a 30mp Bayer array with an AA filter (depending on the strength of that AA filter) I never was overdoing it.
Never thought it was perfect, nor indeed any kind of solution to a problem that generally isn’t there. Most of the time a tried and true and simple Bayer array works fine and doesn’t suffer the light loss stacked photodiodes do.
There is always the argument of better vs good enough, consumers almost universally end up buying ‘good enough’ over better when the differences are hard to discern but there is a price differential.
The fact that some of the 3 Foveon layers pick up light meant for the other layers doesn't have to be considered being "poisoned". If the layer sensors have low noise, then they should be able to have a software 3x3 matrix filter to get the proper R G B values out of them. To show an example of this, the human eye has a green cone and a greenish-yellow cone (instead of a green and red cone). These 2 cones have a QE curve that overlaps each other as much as it differs from each other, but the eye/brain has circuitry that separates the values into what we consider green & red. If there is low noise then circuitry can do the same to eliminate this issue. Now, I'm guessing that the Foveon pixel elements have a lower QE curve and higher noise levels than desired so that would be the reason (I assume) for their lack of success.
If there is enough light available, Foveon sensors are at least theoretically really exciting. They have a higher resolution than a typical Bayer pattern sensor with the same pixel number, are not prone to color moiré, and they are said to deliver beautiful natural colors. Not sure about the latter due to a lack of comparable images (never used a Sigma camera), could also be internet fanboyism. The disadvantage of Foveon sensors is their stack of three color filter+sensor layers, which absorbs a good part of the light before it can reach the lowest red sensor. So, since those sensors are less sensitive, they perform in low much noisier than conventional sensors. Basically it is a trade-off, it's up to user's preferences, those who shoot mostly in the ISO 50-100 range could be quite happy with such a camera.
Maybe they are already fit for it, because they are used to feed a small niche, I always wondered if their camera section is profitable. I wouldn't wonder if not, if their good lens sales have to cross-finance their cameras.
Generally, it only works if the S, M, L curves of human vision's spectral response can be represented as linear combinations of the spectral responses of the sensor's primaries.
Maybe Foveon-like sensors need more than 3 primaries for better color reproduction.
For Foveon, more than 3 primaries would mean a 4th layer. But the 3rd layer is already hamstrung with low red light reaching it due to absorption by the upper layers, so this would be a step in the wrong direction IMHO. And it is possible to use software using occasional dark & light frames with mapping per pixel (linear, logarithmic or look-up-table driven as needed) so that they exactly represent what is needed for human vision when reproduced by monitors. Occasional dark & light frames are already being used by the R5 now.
Foveon arrays will always enjoy better resolution (at the same pixel element size) and lack of moire effects over Bayer arrays due to the fact that all the area of them can sense light in all colors instead of the speckled hit-or-miss pattern of the Bayer array. The problem with Foveon is mainly lower light sensing in lower layers and higher noise than state of the art Bayer arrays. It may be hard for them to ever catch up with Bayer arrays there.
One would need to decrease the thicknesses of the other layers, of course.
Won't change the problem with color aliasing due to non-matching spectral response curves.
If it was easy to reduce the thickness of the layers, I'm sure they would have already done it by now. But as lithography improves the layers might be able to be thinned - we'll see.
I just looked up the Foveon response, here's an interesting DPReview discussion link with graphs: https://www.dpreview.com/forums/post/62546078
If it is accurate (who knows?) then it indicates that there is already a somewhat reasonable QE at red (which I had heard was weaker than G & B). It also mentions the *total* QE of all colors at a few wavelengths. They don't seem to align with the QE of the 3 channels added up together (which leads one to wonder how they assigned those values in the first place?
)If you have known, repeatable pixel element response like this, there is always a way for a sharp programmer to translate the 3 channel values into appropriate R G B values that look like those we are accustomed to seeing from existing camera companies. It just takes fast enough SIMD (single instruction multiple data) hardware with the 16 or so bit data instructions needed to do so. The QE curves do NOT have to match those of the human eye, as long as they give software processed accurate results over the entire range expected from current technology sensors. But these graphs don't have any indication of the noise levels, which others have mentioned as a main weakness of Foveon sensors. If you can't get a clean enough signal or enough dynamic range, then your sensor will never be good enough to compete with current Bayer technologies.
.... regardless of what they use to "color" the filters it is basically impossible to get a response curve that falls exactly in the spectrum they want it too. It may be "red" but I can assure you that it is reflecting to a greater or lesser extent in the yellow, green, and blue areas. In addition, each colorant (dye or otherwise) is unique in that once you decide on one you can't really change or the fingerprint will change..and all of your calcualtions are out the window.
Having said that, it is theoretically possilble to know what those values are that lie outside the target part of the spectrum and in what part of the spectrum they are, which would allow you to build the "perfect" data as a sum of the three layers.
I'm certainly no expert in sensor design (although I am one in color theory), but I suspect the difficulty is in the inherent noise contained in the system and the fact that in some cases the noise signal could very well be greater than the actual signal of the color itself in some areas. In those cases what is color/signal data and what is noise? You can make some base generalizations, and perhaps even establish a baseline of the noise signature of each layer and where it lies. I would think the theoretical issue is that you can't just assume that all of the data below "x" is noise and therefore irrelevant. It could very well have relevant data in it, which makes the mathematical calculations immensely more predictivespeculative vs. accurate.
Having said that, it is theoretically possilble to know what those values are that lie outside the target part of the spectrum and in what part of the spectrum they are, which would allow you to build the "perfect" data as a sum of the three layers.
I'm certainly no expert in sensor design (although I am one in color theory), but I suspect the difficulty is in the inherent noise contained in the system and the fact that in some cases the noise signal could very well be greater than the actual signal of the color itself in some areas. In those cases what is color/signal data and what is noise? You can make some base generalizations, and perhaps even establish a baseline of the noise signature of each layer and where it lies. I would think the theoretical issue is that you can't just assume that all of the data below "x" is noise and therefore irrelevant. It could very well have relevant data in it, which makes the mathematical calculations immensely more predictivespeculative vs. accurate.
I don't know how they currently attach the electrodes to their subpixels, but making more layers may also require switching to a more expensive technology (like stacked silicon-on-insulator). And of course, it increases the required data throughput.
There is. For a color reflection of an incident monochromatic light source with a known wavelength.
The problem is, the coefficients in this translation formula are different for different wavelengths of the incident light, unless the spectral responses of the sensor layers are specially crafted in such a way that it is possible to form the human S, M, L color response curves as their (approximate) linear combinations. Otherwise any such formula will only work with a very specific set of pigments in a very specific light.
Similar threads
- Contains 1 staff post(s)
- Contains 1 staff post(s)
- Contains 2 staff post(s)
- Contains 1 staff post(s)
<-- start Taboola -->