Magic Lantern Improves 5D Mark III Dynamic Range to 14 Stops

[jrista]

multiply d800 pixels with FWC and you get the same as 1dx

(a new question to discuss?)
then please keep your self to the subject, so it not will be a new head room or BSI/FSI discussion

There is no such thing as multiple pixel FWC. Each pixel is independent, and is read out independently. Full Well Capacity is a hardware trait...image size normalization has nothing to do with it. Once the analog signal (as a series of signal values of electric charge) is converted into a digital image (which compresses or expands the analog signal into a fixed digital range...i.e. 10, 12, or 14 bits), FWC is no longer a valid term.

 

[jrista]

And, back to our regularly scheduled programming:

How about those ML guys, eh! Thanks to all the ML guys for figuring this out and giving us an additional option, for those who find the need for it.

 

[tron]

And, back to our regularly scheduled programming:

How about those ML guys, eh! Thanks to all the ML guys for figuring this out and giving us an additional option, for those who find the need for it.

+1 FINALLY ON TOPIC

 

[Drizzt321]

And, back to our regularly scheduled programming:

How about those ML guys, eh! Thanks to all the ML guys for figuring this out and giving us an additional option, for those who find the need for it.

++

 

[ishdakuteb]

why not 36Mp and the best DR ?

your imges look best with a camera that you think it is the best? to me those, i have seen, are not.

D800/E is already 36Mp with 14.5 stop DR... without the ISO tricks

how do you know so sure that nikon does not do iso tricks? i am not so sure about this, but i saw "nikon iso less method" on the first line of implementation part.

//-------------------------

however, kudos to ML team about releasing this engineering test load even though i am not going to put it in my cameras since i do not have under exposed problems like ankorwatt and i am just a hobbyist...

 

[jrista]

I would be curious to see how the ML DR trick works with a more advanced interpolation method. AHD, or Adaptive Homogeneity Directed, is a fairly standard approach, used by LR and Aperture, and is one of the main options in tools like RawThearapy and DarkTable (maybe the default). I wasn't certain, from the PDF, if Alex said he was actually using AHD, or if he was using just the basic interpolation that interpolates all four pixels for each 2x2 intersection.

The reason I'm curious is he did note that cross-interpolating the ISO 100 pixels with the ISO 1600 pixels would be "bad". In any normal demosaicing algorithm, I think that will happen anyway. I know that AHD tries to utilize as much pixel information as it can for each output pixel, so I suspect that if anyone uses this feature in the next ML update, if cross-interpolating 100 with 1600 data produces bad results with AHD...I wonder how anyone will actually be able to use such RAW images if they use Lightroom or Aperture.

 

[Drizzt321]

I would be curious to see how the ML DR trick works with a more advanced interpolation method. AHD, or Adaptive Homogeneity Directed, is a fairly standard approach, used by LR and Aperture, and is one of the main options in tools like RawThearapy and DarkTable (maybe the default). I wasn't certain, from the PDF, if Alex said he was actually using AHD, or if he was using just the basic interpolation that interpolates all four pixels for each 2x2 intersection.

The reason I'm curious is he did note that cross-interpolating the ISO 100 pixels with the ISO 1600 pixels would be "bad". In any normal demosaicing algorithm, I think that will happen anyway. I know that AHD tries to utilize as much pixel information as it can for each output pixel, so I suspect that if anyone uses this feature in the next ML update, if cross-interpolating 100 with 1600 data produces bad results with AHD...I wonder how anyone will actually be able to use such RAW images if they use Lightroom or Aperture.

Hmmm...if I'm reading the PDF right, he's using a basic linear algorithm. It looks like he did some additional work, but I don't think it's using one of the advanced interpolation methods. Want to implement one?

 

[lopicma]

This is some interesting new on two cameras that are years apart in construction.

Didn't the 7D share the same "chip" as the 60D and 350D? ...or am I confusing the "chip" with the "sensor"?

 

[mb66energy]

why not 36Mp and the best DR ?

your imges look best with a camera that you think it is the best? to me those, i have seen, are not.

D800/E is already 36Mp with 14.5 stop DR... without the ISO tricks

how do you know so sure that nikon does not do iso tricks? i am not so sure about this, but i saw "nikon iso less method" on the first line of implementation part.

//-------------------------

however, kudos to ML team about releasing this engineering test load even though i am not going to put it in my cameras since i do not have under exposed problems like ankorwatt and i am just a hobbyist...

then you don't understand the difference in the read out between Canon and Nikon/Sonys sensors

http://pro.sony.com/bbsccms/assets/files/cat/camsec/solutions/E_CMOS_Sensor_WP_110427.pdf
page4

Hmmmmh - if I understand the document you provided right: Sony sensors make three images of different exposures and combine them afterwards. This is possible by a high speed readout (e.g. 75 images per second) and resulting in 25 images per second of high DR images. I think this is sth. like HDR imageing. I don't think that the D800 does sth. like this - it is a little bit different from the surveillance cams described in the document.

Perhaps Canon has integrated it's dual amp but hasn't tried to use it due to patents of Sony ... so it might be possible that Sony uses the alternating ISO readout. The convergence of DR @ high ISOs supports that idea: There is only one amplifier setting available to gain acceptable images so the readout is a "classic" readout with just one amp setting.

 

[Drizzt321]

This is some interesting new on two cameras that are years apart in construction.

Didn't the 7D share the same "chip" as the 60D and 350D? ...or am I confusing the "chip" with the "sensor"?

7D has the same sensor as the 60D, but it has 2 of these ADC chips for faster readout (and faster FPS) rather than 1 like pretty much the rest of the Canon lineup has. Excepting the 1DX, they generally aren't messing around with that or the Cinema line.

EDIT: Sorry, I was thinking of the other, magic lantern dual-iso thread. The 7D has basically the same sensor as the 60D, however it has a dual DIGIC-4 processor chips, while the 60D has a single DIGIC-4 processor.

It's a shame, but Canon would likely dogpile them with lawyers if they did. In fact, I think I remember hearing Canon basically making that statement. I could be wrong about that though.

 

[Krob78]

Hey Neuro, I was wondering if you use or have tried ML on your 7d? If so, what did you think of it? Thanks!

why not, d800 has the same FWC as 1dx but 36 Mp
The best SLR sensor on the market today

I'm sorry, but that is a bold faced lie! The D800 has a full well capacity of 44k, while the 1D X has a full well capacity of 90k. That is a TWO FOLD difference between the two, bub!

Indeed. But let's not allow facts get in the way of the same restated DRivel from the DRoll DRones who DRool over DR. :

D800/E is already 36Mp with 14.5 stop DR... without the ISO tricks

You fell victim to one of the classic blunders - The most famous of which is "never get involved in a land war in Asia" - but only slightly less well-known is this: DxOMark's Scores are useless, biased Bovine Scat.

from the whole sensor area you get exactly the same FWC from 1dx and d800

Full well capacity. You keep using that phrase. I do not think it means what you think it means.

From the whole sensor area? Tell me...what is a 'well' in FWC? How does the 'whole sensor area' affect the number of electons a well can hold?

You stated, "d800 has the same FWC as 1dx." Now you're stating, "from the whole sensor area you get exactly the same FWC from 1dx and d800." I was beginning to think you understood this stuff. I was wrong. See...unlike some, I can admit it. :

 

[neuroanatomist]

Hey Neuro, I was wondering if you use or have tried ML on your 7d? If so, what did you think of it? Thanks!

Sorry, never tried it. I would consider it for my EOS M.

 

[Krob78]

Hey Neuro, I was wondering if you use or have tried ML on your 7d? If so, what did you think of it? Thanks!

Sorry, never tried it. I would consider it for my EOS M.

Considered it for my 7d but could never force myself to actually try it... just some trepidation there. Likewise, I could be a candidate for it on my EOS M... Thank you!

 

[risc32]

i just popped in to see how this was going, and from what i can see the most entertaining aspect of this thread are the sweet "princess bride" lines. neuro, you are the man.

 

[jrista]

why not 36Mp and the best DR ?

your imges look best with a camera that you think it is the best? to me those, i have seen, are not.

D800/E is already 36Mp with 14.5 stop DR... without the ISO tricks

how do you know so sure that nikon does not do iso tricks? i am not so sure about this, but i saw "nikon iso less method" on the first line of implementation part.

//-------------------------

however, kudos to ML team about releasing this engineering test load even though i am not going to put it in my cameras since i do not have under exposed problems like ankorwatt and i am just a hobbyist...

then you don't understand the difference in the read out between Canon and Nikon/Sonys sensors

http://pro.sony.com/bbsccms/assets/files/cat/camsec/solutions/E_CMOS_Sensor_WP_110427.pdf
page4

First, that is an application of Exmor for surveillance cameras, and I do not believe the technique applies for stills photography. The frame rate is a fixed 60fps, which would normally utilize a single long exposure that, depending on the metering, will either underexpose severely to avoid clipping highlights, or overexpose severely in order to avoid blocking shadows. Sony breaks that 1/60th of a second up into medium, short, and very short exposure times which are then HDR blended by hardware, and each single HDR frame stored as a normal video frame. Note, that this only really works when you know your fixed frame interval, and assuming you are able to expose for a long enough time (i.e. it wouldn't work if you needed like 1/320th of a second shutter or something along those lines.)

That is not what the Exmor sensor in the D800 does. The D800 has only a single stage of digital CDS and the on-die column parallel ADC. The use of those two things, while concurrently locating PLL and other high frequency components off in a separate region of the sensor die (to avoid allowing them to introduce noise during ADC), is how Sony and the D800 achieve low read noise for stills.

 

[ishdakuteb]

why not 36Mp and the best DR ?

your imges look best with a camera that you think it is the best? to me those, i have seen, are not.

D800/E is already 36Mp with 14.5 stop DR... without the ISO tricks

how do you know so sure that nikon does not do iso tricks? i am not so sure about this, but i saw "nikon iso less method" on the first line of implementation part.

//-------------------------

however, kudos to ML team about releasing this engineering test load even though i am not going to put it in my cameras since i do not have under exposed problems like ankorwatt and i am just a hobbyist...

then you don't understand the difference in the read out between Canon and Nikon/Sonys sensors

http://pro.sony.com/bbsccms/assets/files/cat/camsec/solutions/E_CMOS_Sensor_WP_110427.pdf
page4

well, like i have said that i do care about technique than camera specs. how about your camera techniques? crap is what i think...

 

[Famateur]

i just popped in to see how this was going, and from what i can see the most entertaining aspect of this thread are the sweet "princess bride" lines. neuro, you are the man.

Agreed!

 

[fman]

Excellent (and very inventive) idea.
Unfortunately the resolution probably will be somewhat reduced, so real improvement should rather come from the sensor.
I hope that Canon will improve the sensor readout noise in future sensors so that there is no need for SW tricks.
Even the full well capacity is bigger for Canon sensors (due to bigger pixels) the readout noise is so high that in case of low ISO it cancels all the advantages of big pixels and limits the dynamic range.

 

[Krob78]

Excellent (and very inventive) idea.
Unfortunately the resolution probably will be somewhat reduced, so real improvement should rather come from the sensor.
I hope that Canon will improve the sensor readout noise in future sensors so that there is no need for SW tricks.
Even the full well capacity is bigger for Canon sensors (due to bigger pixels) the readout noise is so high that in case of low ISO it cancels all the advantages of big pixels and limits the dynamic range.

I hope that Canon will improve the sensor readout noise in future sensors so that there is no need for SW tricks

Need the SW tricks or not, I can only imagine that the software junkies will always pursue improvement and "tricks"...

 

[rpt]

i just popped in to see how this was going, and from what i can see the most entertaining aspect of this thread are the sweet "princess bride" lines. neuro, you are the man.

Agreed!

neuro is the man. As are so many others on this forum.

Why am I on this thread? Never mind!