p.1 #1 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
http://www.canonrumors.com/2013/07/magic-lantern-improves-5d-mark-iii-dynamic-range-to-14-stops/ From Magic Lantern
The Magic Lantern team is letting the world know that they are at the beginning of a new function for the 5D Mark III and EOS 7D. Apparently they’ve figured out a way to boost dynamic range by 3 stops by using a new way of making a photograph by using ISO 100 for some of the vertical lines and ISO 1600 for the rest of it. Sounds wild to us, not sure how they’re doing that. There will be a process to combining the lines of 100/1600 ISO to give you the full resolution file.
From Alex at ML
Works for both photos and raw videos on Canon EOS 5D Mark III. On Canon EOS 7D, photo only (since it can’t do raw video).
What are the benefits? more dynamic range (less shadow noise).
The functionality was not really in Canon firmware, but in the chip that samples the sensor. Canon firmware does not make any use of it (they just set both ISO “registers” to the same value). It was implemented on the hardware side though (and that’s why it works only on these 2 cameras, because only they have two ISO amplifiers)....Show more →
p.1 #2 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
It works, I've used that type of configuration on other sensors.
The main problem is that you DON'T gain 3Ev. The read noise (5Dmk3) at base ISO is about 30e-, and the read noise at ISO1600 is 4e-. For practical purposes they average in quadrature, giving:
RNa = sqrt(RN1^2 + RN2^2) /2 = sqrt(16 + 900) /2 >>> RNa = 15.1e-
That lowering of RN gives a DR increase of 1Ev, the result is 12.1Ev, almost exactly one Ev more than the "original" layout.
Another very big problem with the Canon layout is that you have to pick either the red or the blue column. ONE of them is going to be at ISO1600, seriously skewing the color space available. You can either get bright blue or bright red in the image. Not both. And conversely, only one of the red & blue is going to get a noise advantage, the other is left "as is".
For this you pay an arm and a leg:
The top 4Ev towards white is halved in resolution since you only use half of the pixels, and you also halve the SNR, that is: Less resolution with worse noise. That is everything from middle gray and brighter.
For video this is not a problem, for still images it is a serious drawback.
p.1 #6 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
theSuede wrote:
It works, I've used that type of configuration on other sensors.
The main problem is that you DON'T gain 3Ev. The read noise (5Dmk3) at base ISO is about 30e-, and the read noise at ISO1600 is 4e-. For practical purposes they average in quadrature, giving:
RNa = sqrt(RN1^2 + RN2^2) /2 = sqrt(16 + 900) /2 >>> RNa = 15.1e-
That lowering of RN gives a DR increase of 1Ev, the result is 12.1Ev, almost exactly one Ev more than the "original" layout.
Their algorithm uses the ISO 100 lines, ISO 1600 lines, or a mix of the two depending on the tonal value. For the deep shadows they're using only the ISO 1600 data, which should give them the full read-noise improvement correct? But with half the captured light/reduction in shot-noise SNR since the ISO 100 lines would be skipped.
p.1 #7 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
Unfortunately that kind of algorithm doesn't work at all. Red is only present in "even" numbered columns, blue is only present in odd numbered columns. And to limit the demosaic to twice the amount of sparsity would lower the accurate detail retention ability by 3-4x.
p.1 #8 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
theSuede wrote:
Unfortunately that kind of algorithm doesn't work at all. Red is only present in "even" numbered columns, blue is only present in odd numbered columns. And to limit the demosaic to twice the amount of sparsity would lower the accurate detail retention ability by 3-4x.
The ISO alternates every 2 lines, so the 2x2 Bayer is preserved. Alex @ ML wrote up a whitepaper about it here.
p.1 #9 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
Ah, that makes a difference. Unfortunately that also increases aliasing by quite a bit... Still not very good for still images (if you want full resolution, quarter resolution is the first integer step to remove that noise aliasing in that case).
p.1 #10 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
phuang3 wrote:
There must be a reason why Canon did not use these on-chip function.
When designing the hardware they typically have a list of functions or requirements that they wish to implement into the design, but during the push towards finalization they may choose not to include some things, reasons could be that they needed more R&D to get it to work right but don't have the time to do it before finalization, or they are ment for a future update or new model that uses the same hardware.
Firmwares are similar, in Canon's FW there has been huge chunks of code that were not used because it was only used in another camera body, so rather than rewrite the whole thing, they just abandon what they don't need, and add in what they need.
p.1 #11 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
Well you could just get a d800 or d600 and get over 14ev cheaper without hacks. :-) lol
Sounds pretty cool though the 5diii is a nice piece of equipment
p.1 #12 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
theSuede, do current generation CMOS sensors allow sensels to be read without discharging them? If so couldn't a dual ISO technique be applied against the entire sensor rather than alternating lines?
p.1 #13 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
In theory yes, but most manufacturers do not see the need to do so. So there's no instruction set in the logic circuits to do so... You double the read-out latency, and you only gain about 0.5Ev in a good sensor layout. Some older Exmor sensors had a system that did this to get 14-bit reads from a 12-bit AD ladder. If you mean current CANON sensors in production, I can't give a straight answer...
Most DR-increasing solutions look towards other, more drastic numbers that make the drawbacks seem more worthwhile. Several manufacturers have smaller-format 15-18Ev sensors for sampling/delivery now. Most of the time they work with two physical sections of the drain well in stead of trying to read the same well potential with different amplifications.
p.1 #14 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
'Well you could just get a d800 or d600 and get over 14ev cheaper without hacks.'
There are 90 million EOS lenses out there that will not fit no matter how hard you try.
The original 5D was released in late 2005, it had 11.1 Ev; the 5DIII has 11.7 Ev, a gain of 0.6 Ev in 7 years, or less than 0.1 Ev per year. It sounds negligible because it is.
I have a working hypothesis that Canon do not feel DR is particularly important to still photography. Some of their users agree - I read a poster here state that 'all you need is a different curve.' So maybe that is their thinking.
Sony's RX1 has 14.3 Ev, 2.6 Ev above the 5DIII's measure, at manufacturer's rating of ISO 100 (DxO). If you see the measure as linear, that is over 22%.
[Which by coincidence is the same measure as the gain in resolution the D800 has over the 24Mp cameras, but I believe this DR gain is much more important for non-billboard producers, with respect to final image quality.]
It might be different for others, but as a Sony user I now almost never suffer clipping at the high end, at metered LV. And the colour tonality, and detail at each end of the tone range are very attractive.
High DR (high bit) capture might also make RAW files easier to work in post, and ETTR and HDR are just outdated acronyms for old density workarounds.
p.1 #15 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
philip_pj wrote:
I have a working hypothesis that Canon do not feel DR is particularly important to still photography. Some of their users agree - I read a poster here state that 'all you need is a different curve.' So maybe that is their thinking.
Every time this discussion occurs you need to mentally add "at ISO100" after the word "DR". You have a working hypothesis that Canon do not feel DR at ISO 100 is particularly important to still photography.
You might be right. DR in low light is limited by photon noise, not read noise. Here Canon are more or less level with other FF sensors. If you predominantly shoot at ISO 1600 and above you probably have a very good reason not to care.
I have another theory that they do feel it's important but don't have the ability to do anything about it, due to the requirement for a certain ROI for their sensor fabs.
I'm more than happy with the 5D mkIII for low light work, but I'm hoping for a Sony FF NEX with 14-stop DR high resolution sensor that I can use with my Canon lenses for shooting in good light with CDAF whilst waiting for the next generation sensor from Canon.
p.1 #16 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
That is indeed true. As soon as you pass ISO400 in the Canon camera, things level out (compared to other manufacturers).
The main problem is that given good enough lenses, the D7100 outperforms the 5D3 and the 1Dx for static image quality at ISOs 100 and 200. Which is why some people feel quite snubbed, I guess.
p.1 #17 · Magic Lantern to boost 7D & 5DIII DR by 3 stops!
Matt Grum wrote:
Every time this discussion occurs you need to mentally add "at ISO100" after the word "DR". You have a working hypothesis that Canon do not feel DR at ISO 100 is particularly important to still photography.
You might be right. DR in low light is limited by photon noise, not read noise. Here Canon are more or less level with other FF sensors. If you predominantly shoot at ISO 1600 and above you probably have a very good reason not to care.
