Previous news story    Next news story

Magic Lantern firmware boosts dynamic range of Canon 5D III, 7D

By dpreview staff on Jul 17, 2013 at 19:36 GMT

The folks at Magic Lantern are no stranger to adding new features to Canon DSLRs, courtesy of their EOS Camera Tool software. Their latest creation - called Dual ISO - dramatically increases the dynamic range of the 5D Mark III and EOS 7D by four stops, bringing the total dynamic range to 14EV. This allows you to pull detail out of the shadows with a lot less noise than with the 'stock' firmware. The technical details are complex, but simply put, it works by interlacing two rows of pixels captured at ISO 100 with two rows taken at a higher sensitivity (usually ISO 1600). The low sensitivity rows capture highlight detail, while the high ISO rows capture shadow detail. When they are combined, you get relatively noise-free shadows without blowing highlights.

According to Magic Lantern, there are some downsides to using the Dual ISO firmware, though. Vertical resolution is reduced by half and there's more moiré and aliasing in over and underexposed areas. The author also warns that since this software modifies the sensor's operation, you could end up frying your camera. If you're feeling brave, you can find details on how to install the software here.

Dual ISO works for both stills and movies on the EOS 5D Mark III and for stills only on the EOS 7D. 

This example shows good shadow detail in the foreground, while the highlights in the windows are not clipped. Unfortunately, the author doesn't provide a 'before' image, so it's hard to see exactly what's improved.

If the example above was filmed at ISO 100 to preserve highlights (in other words, underexposed), the foreground would be dark. If you tried to bring up the shadows in post-processing, noise levels would be very high. Filming at ISO 1600, on the other hand, brightens the foreground but clips the highlights in the windows. By combining these two images, you get the best of both worlds.

Via: PetaPixel, Source: Magic Lantern

Comments

Total comments: 66
Severn Bore
By Severn Bore (9 months ago)

Is this hack already available. If so, has anyone used it in the real world and with what results?

0 upvotes
klehocz
By klehocz (9 months ago)

One problem I see is that only every second line has red and blue sensor sites - if you amplify odd/even lines differently, you will only get the increased DR with one of the color components missing. (at least in theory, I haven't looked at this closer)

0 upvotes
klehocz
By klehocz (9 months ago)

So, to clarify... the extra highlight space that this would give would only be for R+G or for B+G. Whether to amplify the odd or even lines should probably be decided on a scene by scene basis (eg. based on the live histogram). It might be of practical use for a single f-stop of enhancement. Could be useful eg. for concert photography, maybe sports. (Unsure about eg. landscapes, because the demosaic algorithm will have a harder time to resolve detail - as noise level changes from line to line.)

0 upvotes
Rachotilko
By Rachotilko (9 months ago)

I don't get it. By definition, you can't add any DR to camera by firmware update, as it's given by the noise levels of the sensor in the shadow areas.

In other words, this is JPEG tweak, it does not translate to DxOMark results, since these are RAW based.

0 upvotes
Revenant
By Revenant (9 months ago)

This is not a JPEG tweak, since the dual signal amplification happens before demosaicing. Just like any other processing, or "cooking", of raw files before conversion, it would affect DxOMark's results. However, DxO usually indicates this in their graphs, when they are aware of it. See, for example, some Pentax and Nikon 1 cameras.

0 upvotes
Ferling
By Ferling (9 months ago)

Interesting. So it's a live mid-tone contrast boost or (for those of you familiar with LR3), a 'Fill' light tool. Nice.

0 upvotes
Ferling
By Ferling (9 months ago)

Update. Further reading suggests users having issues with heavy color moire and stair-stepping on edges and highlights in stills. It would depend on your subject matter and the kinds of output that would negate that. Journalistic with no flash, and maybe weddings might benefit, (would have to test and examine some prints to know for sure regarding weddings and commercial use). Gonna wait this out, and might play with it when the warranty expires on one of my bodies.

0 upvotes
naththo
By naththo (9 months ago)

How about I did still raw image of Canon 7D on Raw Therapee and used no lighting enhancement and picked out the color profile of Canon 7D enhanced Photivo ICC profile and dynamic range is much wider than Photoshop lightroom! So I am thinking why can't Photoshop Lightroom do the same?

0 upvotes
Jurka
By Jurka (9 months ago)

Sony A99 DRO feature do the same. Without any hacks or lanterns...

0 upvotes
zodiacfml
By zodiacfml (9 months ago)

Little advantage over RAW post processing or camera maker's proprietary DR enhancements.

0 upvotes
lkent18
By lkent18 (9 months ago)

Dope... had this idea several months ago. Maybe I should have tried to sell it :-).

0 upvotes
lkent18
By lkent18 (9 months ago)

Well, it was a similar idea, but mine was better :-)

0 upvotes
Artistico
By Artistico (9 months ago)

And Fujifilm had the same idea 15 years ago...

0 upvotes
Henrik Herranen
By Henrik Herranen (9 months ago)

One thing worth mentioning is that vertical resolution is halved only on the brightest and darkest areas where there is no tonal overlap in the two half-images. Midtones are at full resolution. And, if you use a less dramatical setting than ISO100/1600 (e.g. ISO100/400), the full-resolution overlap area is larger. Perhaps DPreview would like to correct this in the news blurb?
It's all thoroughly explained in Magic Lantern's documents, with graphs and all.

Comment edited 2 times, last edit 2 minutes after posting
3 upvotes
wfektar
By wfektar (9 months ago)

Good points, both of them.

It's also fair to point out, however, that if you need the 14 EV, it's probably because you're trying to bring out something in the highlights and shadows, beyond what can be done at native resolution. And that the 100/400 combination won't get you 14 stops. This would be a very ingenious approach to get 14 EV if it had not already been demonstrated that you can get that at full resolution.

0 upvotes
mini23
By mini23 (9 months ago)

Come on DPR guys - I'm not a native English speaker but even I know approximately how to use apostrophes - what is sensor's supposed to mean?

...this software modifies the sensor's operation...
means as much as
...this software modifies the sensor is operation...

1 upvote
Robert98
By Robert98 (9 months ago)

It's a possessive apostrophe - the operation "belongs to" the sensor, so it's the sensor's operation. Your example shows a different use of the apostrophe, to indicate a missing letter.

21 upvotes
jadot
By jadot (9 months ago)

Native english speakers have learned how to use the possessive as well as the omissive. It's the correct use.
The internet is rife with pedantic spelling and grammar policing, and I'm all for it, but it's now got to the stage where people are getting upset at the mere appearance of an apostrophe.
Often, re-reading what is before you can save you from making an embarrassing mistake, grammatically speaking.

15 upvotes
Sam VG
By Sam VG (9 months ago)

German attaches the genitive suffix without an apostrophe, which may be the cause of your confusion. In German you write: "Martins Frau". In English you write: "Martin's wife".

0 upvotes
Lajos Hajdu
By Lajos Hajdu (9 months ago)

In this case,"sensor's" means "of the sensor". mini23, you should learn basic English before erroneously criticizing others. (Also, what you propose - "this software modifies the sensor is operation" - is totally gibberish.)

Comment edited 2 minutes after posting
1 upvote
reality_check
By reality_check (9 months ago)

Apart from the indiscriminate use of the dash (and dots) that you seem to be a proponent of, you are telling us that the use of the apostrophe here is erroneous. Sadly, you are wrong. While the English spelling system is totally flawed and erratic (and really should be reformed), this is good writing. Sorry, Mini!

0 upvotes
yabokkie
By yabokkie (9 months ago)

maybe +4 stops dynamic range from 70D at no cost of resolution,
but at the cost of near 1 stop worth of SNR ?

Comment edited 2 times, last edit 6 minutes after posting
0 upvotes
Photog74
By Photog74 (9 months ago)

Sounds like an excellent trade-off to me.

2 upvotes
yabokkie
By yabokkie (9 months ago)

exactly. 1 stop worse SNR in highlight is not a big deal, especially at low and mid-range ISO settings. for high ISOs underexposure works well now (in a different direction though) and there is no difference between Canon and Sony sensors for ISO > somewhere between 800 and 1600.

Comment edited 10 minutes after posting
1 upvote
planetMitch
By planetMitch (9 months ago)

This update from Magic Lantern will not work on anything but the 5D3 and the 7D. Our original report (which petapixel used as the original source and I wish DPR had used me as the source :) shows it only works on these 2 cameras -- http://blog.planet5d.com/oylb

2 upvotes
yabokkie
By yabokkie (9 months ago)

please do explore possibilities of 70D as soon as you could, from dynamic range to 3D.

0 upvotes
Ferling
By Ferling (9 months ago)

Noise is not so apparent if you expose to the right and the entire scene is evenly lit, (I have a good number of wedding shots to back it up). Those extra stops would mean I could use f4 (more folks in focus), or an increase in shutter speed (especially on at 300mm) to stop more motion.

0 upvotes
Adam2
By Adam2 (9 months ago)

How does this differ radically than say doing a HDR with two images? Obviously, it might be better when shooting moving subjects but otherwise, what's the point?

1 upvote
yabokkie
By yabokkie (9 months ago)

at least it's an ideal tool for moving subjects when we get blurred image most of the time (either shutter or focus).

0 upvotes
makofoto
By makofoto (9 months ago)

Magic Lantern is mainly used for video, ie. no HDR

0 upvotes
planetMitch
By planetMitch (9 months ago)

Magic Lantern is used for all kinds of photos - and has some great HDR functions as well as things like an intervalometer built in (which I still can't believe Canon doesn't include)

3 upvotes
AbrasiveReducer
By AbrasiveReducer (9 months ago)

I have a 5D3 but when I use the built-in HDR function I see edge artifacts. So the ability to get it all in a single Raw image instead of 3 jpgs sounds good to me. The part about "may fry your camera" doesn't sound that good, though.

Comment edited 26 seconds after posting
0 upvotes
solarider
By solarider (9 months ago)

Re: "Vertical resolution is reduced by half and there's more moiré and aliasing in over and underexposed areas. The author also warns that since this software modifies the sensor's operation, you could end up frying your camera."

Yikes.

Simple, for 14.1 stops get ye the Pentax K5, K5II, of K5IIs, pull all you want from the shadows and forget about all of the ML downsides.

Best Wishes

5 upvotes
Henrik Herranen
By Henrik Herranen (9 months ago)

The vertical resolution thing is incorrect. Resolution is not lost in all of the image, "only" in the brightest and darkest areas. At midtones where both of the two half-images contain tonal information, full resolution is available.
Also, about the "frying" part: that's a disclaimer they need to say just in the unlikely case the modification breaks something. However, just setting two ADC amplifiers to different positions doesn't sound too likely to break anything.

Comment edited 2 times, last edit 2 minutes after posting
3 upvotes
Zerblatt
By Zerblatt (9 months ago)

Or for a real kamera Nikon D800 14.3 Evs

0 upvotes
angarhurta
By angarhurta (9 months ago)

...Neither the D800 nor any Pentax record video with 14 f-stops of dynamic range, because the video mode in both systems is highly compressed, so the dynamic range barely reach a 11 f-stops or so of dynamic range... so even the 12 f-stops of dynamic range in the Canon 5d mark iii raw video mode is far far far superior... No doubt the canon 5d mark iii is´╗┐ the best camera for a photo/video-grapher; It takes photos with a nice resolution, it has the best auto-focus system ever developed, it's a great low light performer, can record full uncompressed 14 bit raw video, it takes photos an records raw video with 14 f-stops of dynamic range, it's well built, etc...

0 upvotes
Mr Blah
By Mr Blah (9 months ago)

Just an FYI: This doesn't seem to be available for use with the regular video modes on the 5D Mk. III or 7D. It only works in tandem with the previous raw video/time-lapse hack or with regular raw photographs.

Comment edited 2 minutes after posting
2 upvotes
ImagesToo
By ImagesToo (9 months ago)

I think this is not correct and doesn't do anything. You get ISO1600 by amplifying the signal which the sensor captures. You don't get any better noise performance because the noise is the square root of the number of photons in a pixel and you haven't changed that. You might just as well amplify the ISO100 signal (ie increase the brightness of the shadows after processing) and you will get exactly the same result without all the moire. All it is doing is effectively a really crude hdr processing on an image and compensating for the poor RAW to jpg conversion in the camera. Bracketing exposures and using hdr will give better results.

1 upvote
Henrik Herranen
By Henrik Herranen (9 months ago)

It is actually you who are not correct.
The thing that is compensated for is the noisy ADC (analog-to-digital converter), which follows after the ISO amplifier. There is nothing wrong with 5D3's and 7D's RAW handling; the error happens at the ADC before ever getting to the RAW stage. That is what the new modification helps with.
As for bracketing exposures: it is all good and fine if you have a non-moving subject. But Magic Lantern's page specifically tells that it addresses the situation where there is movement. In such cases multi-image methods naturally fail.

Comment edited 2 times, last edit 13 minutes after posting
2 upvotes
ImagesToo
By ImagesToo (9 months ago)

I agree about the amplifier being before the A/D will obviously help with A/D noise. Of course the question remains just how noisy is the ISO amplifier itself because you obviously can't do anything about that. In general a 14 bit A/D has to have a noise level well below 14 bits else it can't possibly give 14 bits of conversion and the same applies to the ISO amplifier. No matter what you are still scratching around in the deep shadows where the quantum noise starts being significant. Actually even with a moving subject you can sometimes use bracketing but it becomes messy and you have to do it in small segments of the image where you can get alignment. A single raw capture processed using hdr usually works pretty well though without loss of resolution.

0 upvotes
ImagesToo
By ImagesToo (9 months ago)

Further comments: A typical high end sensor has a capacity of around 65000 electrons (photons). A 14 bit A/D has a dynamic range of 2 to the power 14 = 16384 discrete levels with each level thus having an effective difference of 4 electrons. At the dark end you finish up with 4 electrons in a bit with a quantum noise level of +-2 electrons (ie the signal to noise ratio is 1:1) which is so bad as useless. To be useful the amplifier has to have an input measured noise level of well under 4 electrons and even then it can't do anything about the very low quantum S/N ratio at the dark end. That's a huge challenge for amplifier design and we are ignoring the effects of thermal noise on amplifier performance.
The amplifier will raise the signal level to the A/D converter so it's noise performance isn't going to be the dominant factor in any case. So the limiting factor is going to be the amplifier and quantum noise neither of which you can do anything about.

0 upvotes
Sergio Rojkes
By Sergio Rojkes (9 months ago)

...wake me up when upgrades stop, please...

0 upvotes
JWest
By JWest (9 months ago)

Nighty night!

0 upvotes
AbrasiveReducer
By AbrasiveReducer (9 months ago)

Wow. If this is true, the one area where Canon is significantly behind Nikon is fixable.

2 upvotes
AngryCorgi
By AngryCorgi (9 months ago)

As long as a 4.5MP 7D or a 5.5MP 5D3 is acceptable to you.

12 upvotes
Thanatham Piriyakarnjanakul
By Thanatham Piriyakarnjanakul (9 months ago)

Do it at your own risk, only rich kid will do that.

2 upvotes
Henrik Herranen
By Henrik Herranen (9 months ago)

AngryCorgi: DPreview misrepresents the resolution issue. Resolution is halved only at the brightest and darkest areas. And only vertical resolution, not horizontal. Midtones are presented at full resolution.
Thanatham: Magic Lanters has always had the disclaimers there, nothing new there. Still many many people have used it during the last years.

7 upvotes
yabokkie
By yabokkie (9 months ago)

it says the resolution is halved, not quartered. it's sad that Canon got problem with both dynamic range and sensor resolution or we may be able to beat D800 with a 60MP 5D3'.

0 upvotes
mr.izo
By mr.izo (9 months ago)

hmm, 7d only for stills? hope they're working on 6d too, or it's already integrated?

0 upvotes
Revenant
By Revenant (9 months ago)

Apparently, only 5D3 and 7D have two signal amplifiers, which is necessary for using two ISO values simultaneously.

4 upvotes
mr.izo
By mr.izo (9 months ago)

so, no need for d800 any more, hehe? if this new feature can solve shadows etc like mentioned nikon, than 6d is more an more option for me..

1 upvote
vincent__l
By vincent__l (9 months ago)

As with the new Aptina Clarity+ sensor just announced, Sony's BSI sensors, Foveon sensors, Fuji's X-Trans, and most likely Canon's new dual diode phase detect sensors, there's no free lunch. It's a trade-off. You will improve something while sacrificing something else that is hopefully less important to you.

11 upvotes
JJ Rodin
By JJ Rodin (9 months ago)

Yes, that 'physics' thing getting in the way again! :(

You can engineer around limits but physics is the master of all things, that BA$TURD.

Fool the brain is what it is all about, politics and photography ! :)

0 upvotes
iAPX
By iAPX (9 months ago)

Great work, and incredibly interesting paper, I hope they patented it because this is something that should have been implemented in any modern sensor to enable very high dynamic capture!

It means that many actual sensors might be used in extended HDR mode, not actually faking HDR by changing the curves, but with real dynamic extension. Very interesting!

The most interesting part is that could also be applied to compact-size sensor or even smartphone if they have 2channels to read their sensors! ouch!

Comment edited 1 minute after posting
0 upvotes
liviutza
By liviutza (9 months ago)

Unfortunately channel readout is not the only involved part - you also need a dedicated controller line able to set EV (or ISO) at a different level every other pixel. Doable (see Fuji), but not a must for Bayer sensors hence not present in most of them.

3 upvotes
mpgxsvcd
By mpgxsvcd (9 months ago)

I wish they had shown the full size image. It has horrible moire and the resolution really is cut in half.

My Panasonic GH3 does the exact same thing in video and stills. It is called the High Dynamic mode and it really isn't that great.

First off it doesn't increase the dynamic range. It simply compresses the dynamic range of the scene you are capturing. Those are two totally different things.

I applaud their efforts. However, the actual results are far worse than what is illustrated here.

6 upvotes
imsabbel
By imsabbel (9 months ago)

Actually, those two are NOT different things, as long as you operate in RAW (as the 14bit available are not even remotely used by the native sensor SNR)

Of course, you WILL need to compress the dynamic range for any kind of output medium (be it print, which only has somthing like 100:1, or screens where even the best top at something like 2000:1)

Also, while the moire is crap, thats not unfixable. You would just need to write an optimized raw decoder taking that into account. Different ISO per scanline is much less bad an artifact that the basic baeyer pattern, in terms of reconstrution..

Comment edited 1 minute after posting
6 upvotes
vincent__l
By vincent__l (9 months ago)

It looks like this is a feature that Canon has implemented in hardware but disabled in firmware. I'm wondering how Canon will react to this given that they often use firmware to differentiate their product line.

0 upvotes
tkbslc
By tkbslc (9 months ago)

If they are just switching ISO every other frame, then it is not something special Canon would have had to hide.

0 upvotes
vincent__l
By vincent__l (9 months ago)

ML claims that the feature, which requires "dual ISO amplifiers", is implemented in hardware but disabled by Canon. They also claim that only the 5D3 and 7D have this feature. Seems like Canon has been experimenting with it but were not ready to go into production.

Source:

http://www.canonrumors.com/2013/07/magic-lantern-improves-5d-mark-iii-dynamic-range-to-14-stops/

Comment edited 2 minutes after posting
3 upvotes
NetMage
By NetMage (9 months ago)

I'm not sure it is an experimental feature so much as a consequence of the multiple readout channels that require separate amplifiers in order to provide faster performance. What we normally consider a problem (scan lines in the picture slightly differently amplified if the amps aren't tuned identically) ML has turned into an advantage.

3 upvotes
Mr Blah
By Mr Blah (9 months ago)

@tkbslc
It's not switching ISO each frame (though that can also be enabled via a different part of the Magic Lantern firmware). Instead, it's exposing different pixels on the sensor at different ISOs at the same time. The post-production workflow interpolates an even exposure for each image, resulting in complete frames.

2 upvotes
liviutza
By liviutza (9 months ago)

Fuji experimented with this idea in their HR sensors (photo receptors with different sizes) and EXR sensors (similar receptors with different exposures), and their DR performance has always been great - hell, the S5 PRO was about 5 years ahead of its time in this respect. The idea does require some special hardware as well as processing, and one must have great control over the manufacturing and development pipeline in order to do it - Nikon / Pentax for example would be limited by whatever Sony decides to do with the sensors, but Canon does have some freedom in this respect. It would be great if this feat were technically possible with the sensor inside the 70D (no idea, but that's built with dual receptors for each pixel). Anyway. Canon not working on it themselves or at least officially supporting ML is quite a disappointment.

4 upvotes
yabokkie
By yabokkie (9 months ago)

70D also got 8 channel readout like 7D and 5D3 and might be hacked similarly. but since single exposure HDR is an obvious benefit of the new sensor design, hopefully we will get it someday as standard Canon feature.

Comment edited 3 times, last edit 7 minutes after posting
1 upvote
Bill Bentley
By Bill Bentley (9 months ago)

I supported Alex & Co. for enhancing my 600D. I'm looking forward to supporting them again after I upgrade to the 70D. Keep up the good work boys.

1 upvote
Gesture
By Gesture (9 months ago)

Didn't Olympus E-10/E-20 have something similar for boosting ISO. But one would think we'd have seen more ideas like this.

0 upvotes
Total comments: 66