July 17, 2013

Magic Lantern Comparison: Canon 5D Mark III Normal RAW vs. 14 Stop Dual ISO RAW

Video thumbnail for vimeo video First Magic Lantern Comparison: RAW vs. 14 Stop Dual ISO RAW - No Film SchoolYesterday we had the big news about Magic Lantern's big breakthrough: the ability to actually shoot with two different ISOs on the sensor at once. This is quite a technological leap from the previous hacks, as it is actually modifying how the sensor operates at a very base level. We've got a brief comparison from Michael Anthony showing off what the new hack can do, and we've also got some RAW Dual ISO samples from Luke Neumann for you to play around with.

Here are the settings used by Michael Anthony in the test:

Camera: 5D Mark III (shot 24fps at 1080p)
Dual ISO was set to use 100 and 1600 ISO.
Footage was shot in a dimly lit room with both a tungsten lamp and dark areas in the frame.
Shot handheld to reveal any moire or aliasing which may increase using Dual ISO feature.
Raw was converted to DNGs using the newest rawtodng.exe.
Postprocessing via Adobe Camera Raw using identical settings for both Normal Raw and Dual ISO shots...
Exposure was brought up a bit to +2.8
Highlights -100,
Shadow +70,
Sharpening 0,
Noise Reduction 0,
All other settings were at 0 or default values
Exported via After Effects to intermediate AVI files (Cineform 444 Filmscan 1)
Edited in Premiere Pro CS6 and rendered to H.264 1080p (10Mbps).

http://vimeo.com/70459941

[Update]: Michael recently uploaded a new video:

I've had a number of people ask how raw video "Denoised" would fare compared to the results obtained using the Dual ISO feature. So here's another test comparing exactly that.

Ran a moderate level of denoising using Neat video. I tried to retain as much detail as possible to prevent the footage from becoming too plasticky.

http://vimeo.com/70511941

This is one shot of the sample video shot by Luke in anamorphic:

From the samples I've seen so far there are still some issues to work out. There will likely be aliasing and moire no matter what you do, but different debayering methods can help. If you've read a1ex's PDF, you'll know there is a lot of really complicated calculations going on in order to compensate for the fact that the camera is shooting two different ISOs at once. As I said before, I think this will be useful for special cases when you are dealing with really tough contrast ratios, not necessarily something you would use on a daily basis. Things could improve, and it's best to keep that in mind with any of these hacks -- especially since there has been almost no testing and the code is in the very early stages.

You can download Luke's sample footage here (it's anamorphic so you will have to de-squeeze it a bit).

Links:

Your Comment

53 Comments

It's a 1.5x lens for those interested.

July 17, 2013 at 3:13PM, Edited September 4, 11:21AM

0
Reply

What lens exactly?

July 17, 2013 at 3:36PM, Edited September 4, 11:21AM

0
Reply

Kowa 1.5x Anamorphic attachment on an 85mm Jupiter f 2 from Vid-Atlantic.

July 17, 2013 at 3:47PM, Edited September 4, 11:21AM

3
Reply

Both look like poo

July 17, 2013 at 3:14PM, Edited September 4, 11:21AM

2
Reply
Joe

I have to agree... but has to start somewhere. Really looking forward to what develops.

July 17, 2013 at 3:21PM, Edited September 4, 11:21AM

5
Reply
5d3 user

Long ways to go. You're resolving less than 720p on these cameras. As faster CF cards come out we should be able to shoot at higher resolutions which should improve this function. As of right now it's much more useful in photo mode. Pretty easy/painless way to take HDR photos.

July 17, 2013 at 3:26PM, Edited September 4, 11:21AM

2
Reply

haha yeah sorry, definitely looking forward to more developments!

July 17, 2013 at 3:30PM, Edited September 4, 11:21AM

0
Reply
Joe

"resolving less than 720p"? That was said to be true of the camera's native H264 encoding, possibly due to whatever awful scaling method it employed, but I'm definitely seeing a massive improvement in detail in my 1080p24 5D3 raw video tests over the original captures. It'd be interesting to see an actual resolution test, with objective numbers. I suspect magic lantern raw video on a 5D3 at 1080p is getting around 900 lines of horizontal resolution.

July 18, 2013 at 6:59PM, Edited September 4, 11:21AM

0
Reply

Update: found this Canon 5D3 ML raw 1080p vs standard h264 1080p resolution test:
https://vimeo.com/67090302

July 18, 2013 at 7:42PM, Edited September 4, 11:21AM

0
Reply

I was talking about in HDR mode where you're only resolving half of the vertical resolution.

July 29, 2013 at 7:45PM, Edited September 4, 11:21AM

6
Reply

With HDR that might be true, but the dual ISO processing merges the images so that you receive (up to) full resolution in middle tones, while highlights and dark shadows only use one of the two images, thus half the resolution (but still better than clipping, which basically means no data at all!)... really clever math behind that - I checked it out just today! :)

I fear the low resolutions are not due to the card speed (looking at what the 1DC records on CF cards!!) - pretty much all of Canon's DSLRs should be capable of (relatively compressed) 4K, and I find it quite upsetting that we're just not getting cameras that do all they could. For me the eye opener was the Nikon 1 that can shoot one-and-a-half-second sequences in 4K that look really beautiful - with that "crappy" processor!! Now imagine what a 5D with its Digic 5+ processor should be capable of. I am still disappointed that the 1080p image is not actually a down-sampled full-size image... well now you know the painful inside of my head! ^^

October 12, 2013 at 6:30PM, Edited September 4, 11:21AM

1
Reply

The aliasing in the dual ISO shot is really ugly. I don't know which one is more distracting, noise or aliasing...

July 17, 2013 at 3:20PM, Edited September 4, 11:21AM

0
Reply
Bry

Very ugly aliasing indeed, and purple color artifacts mean bad demosaicing. But hey, one line out of two, two different exposures... what can you do? There will certainly be better ways to interpolate the missing lines, but personally, I wouldn't trade sharpness for dynamic range. The 10+ stops from the 5D MkIII raw footage are enough for me to work with, and if I want both sharpness and DR, I'd rather direct my attention to the BMCC, which is cheaper anyway.

I just wonder how he managed to get so much noise on the first video... Other 5D Raw footage on the web doesn't seem to have this issue (?)

What is fascinating is that deep sensor technology color science is going open-source and that's pretty awesome.

July 17, 2013 at 3:33PM, Edited September 4, 11:21AM

11
Reply

I think the noise is due to the post-exposure applied to the ISO 100 material.
That's obviously much more of an issue than I suspected - then again if it wasn't, the whole point of dual ISO would be void.
If that wasn't an issue, we could simply shoot everything at ISO 100 (and very short exposure), and display it on monitors as bright as the sun, achieving a perfectly authentic real life dynamic range. Anyone ever thought about that: The lack of dynamic range in our monitors and on paper is why we have to play so much with dynamics and exposure while recording the image - that, plus we don't have sufficiently high-res file formats that can capture that dynamic range at a proper resolution (not pixels, put shading stops)...

October 12, 2013 at 6:40PM, Edited September 4, 11:21AM

11
Reply

I don't do much in the way of dslr shooting these days so I could me wrong but doesn't all that noise in the raw sample make it basically unusable?

July 17, 2013 at 3:34PM, Edited September 4, 11:21AM

0
Reply
Jerome (also..b...

You trade noise in the shadows for aliasing and loss of resolution in the shadows (and probably highlights as well; hard to see that in this clip). Makes you wonder if aggressive noise reduction wouldn't be a comparable solution when motion is mild enough.

Just an observation. Good start. Keep it up.

July 17, 2013 at 3:37PM, Edited September 4, 11:21AM

4
Reply
NashDP

I just uploaded an additional comparison of Normal raw video "Denoised" vs Dual ISO raw video. https://vimeo.com/70511941

July 17, 2013 at 5:21PM, Edited September 4, 11:21AM

8
Reply
Michael Anthony

If the sensor is capable of 14 stops dynamic range, why can't each individual pixel behave like this? Is it because the enormous bit depth would create too large files that Canon has split it's raw files into iso-ranges? How is it done in the Alexa?

July 17, 2013 at 3:41PM, Edited September 4, 11:21AM

0
Reply
Eric

The Canon sensor 5d3 is ratted at 11.7 stop DR in raw mode. Not every camera is the same even in Raw, like the Nikon D800 which is at 14.5 stop.

July 17, 2013 at 5:03PM, Edited September 4, 11:21AM

7
Reply
Daniel

@Daniel, thanks for your reply. I do realize that sensors differ in dynamic range. But a specific sensor can vary in sensibility (100/1600 iso). So the sensor/pixel is capable of wide dynamic range, but – as it seems – just not in one frame/picture! Why is that?
The technique of 'dual gain' – that has made the Alexa the most successful cinema camera – is here. So it should be possible to use it on the mkIII.

July 17, 2013 at 5:42PM, Edited September 4, 11:21AM

0
Reply
Eric

Not trying to sound derogatory, but if it were that easy wouldn't all cameras be doing it?

July 17, 2013 at 5:59PM, Edited September 4, 11:21AM

0
Reply
Brian

Of course, but I'm interested in the factors that make it so difficult.

July 18, 2013 at 1:12AM, Edited September 4, 11:21AM

0
Reply
Eric

@Daniel, thanks for your reply. I do realize that sensors differ in dynamic range. But a specific sensor can vary in sensibility (100/1600 iso). So the sensor/pixel is capable of wide dynamic range, but – as it seems – just not in one frame/picture! Why is that?
The technique of 'dual gain' – that has made the Alexa the most successful digital cinema camera – is here. So it should be possible to use it on the mkIII.

July 17, 2013 at 5:42PM, Edited September 4, 11:21AM

4
Reply
Eric

The Alexa has hardware that can dual read the same pixels at different gain levels...Canon cameras can't do that, so ML is only gaining up every other line of pixels and combining the lines. That's why there's a resolution hit and you get nasty jaggie artifacts.

July 17, 2013 at 6:35PM, Edited September 4, 11:21AM

0
Reply
Gabe

I'm wondering the same thing as Guillaume. Why is the original raw image so noisy? Was the ISO cranked very high? Does shooting raw on the 5D cause more noise in the image?

July 17, 2013 at 3:44PM, Edited September 4, 11:21AM

7
Reply
Rick

I was the one who shot the Dual ISO test video. The reason the original raw image is so noisy is that I shot it at 100 iso to preserve the highlight detail of the lamp. I brought up the the shadows in Adobe Camera Raw to where I could see information and this was the result. The test doesn't really show how dark the original scene was. In addition to demonstrating the Dynamic range improvements, the purpose so the test was also to stress the new feature and reveal it's weaknesses such as aliasing. The tech document written by Alex of Magic Lantern indicates that the interpolation algorithm he used is far from optimal, and that better results can likely be achieved in the future using better interpolation methods. Remember that raw video, a fairly recent development, started only supporting a few FPS and now is completely usable at 24fps and above. In fact I just returned from my last commercial gig and used raw exclusively and it worked flawlessly. I far as I'm concerned, this Dual ISO feature just gives us another tool to use when the situation calls for it.

July 17, 2013 at 4:11PM, Edited September 4, 11:21AM

1
Reply
Michael Anthony

Thanks for the test Michael. Just had to say that

July 17, 2013 at 9:19PM, Edited September 4, 11:21AM

7
Reply
Archie

Thanks Archie.

July 17, 2013 at 11:37PM, Edited September 4, 11:21AM

0
Reply
Michael Anthony

I don't know if anyone else is getting this ugly purple color on the dark spots of the image. (the shadows) + the aliasing is really bad. Great start though. lol

July 17, 2013 at 5:05PM, Edited September 4, 11:21AM

0
Reply
Oscar

The decreased resolution makes this a dealbreaker for me. I disabled it until they can figure something else out.

July 17, 2013 at 5:10PM, Edited September 4, 11:21AM

0
Reply
Harry Pray IV

Man, that dual ISO video kinda looks like it's had a boatload of denoiser applied to it - it's got that very muddy, almost vector-ish look to it. Not saying that the guy who shot the test is being dishonest, I'm just making an observation on the result.

July 17, 2013 at 5:42PM, Edited September 4, 11:21AM

12
Reply
Brian

I didn't apply any denoiser in that test. I also did a separate test comparing raw video "denoised" to Dual ISO raw video. https://vimeo.com/70511941

July 17, 2013 at 6:07PM, Edited September 4, 11:21AM

0
Reply
Michael Anthony

Thanks for re-posting the video de-noised. It's easier to compare the two. I don't see much, if any, increase in dynamic range.

July 17, 2013 at 6:44PM, Edited September 4, 11:21AM

0
Reply
Rick

Guys, chill out, this is early alpha software testing not a production ready function - its an experiment to show the potentials. To claim this is unusable at this early stage is really a joke, off course there will be noise, moire - why don't you get of ass and try to do something as groundbreaking as this?. Get over it and get real!

July 17, 2013 at 5:57PM, Edited September 4, 11:21AM

0
Reply
shaun wilson

It still looks like DSLR footage.

Just get a GH3 if you want better video. Save the money and patching. Stick, though, with the Canon over Panasonic for still photography--or better yet get a Fuji X-E1 for stills photography.

..... or........ just keep patching, and patching, and patching, and patching, and patching, and patching....

July 17, 2013 at 8:15PM, Edited September 4, 11:21AM

2
Reply
Gene

dang, meant to say Fuji X Pro

July 17, 2013 at 8:18PM, Edited September 4, 11:21AM

4
Reply
Gene

Gene, every single post of yours is a joke and pro GH3..if you stopped being a troll, and had both, which I do, you would soon realize that the GH3 is as far from a 5d3 raw as a 5d3 is from the Alexa, cheers.. :)

July 17, 2013 at 8:39PM, Edited September 4, 11:21AM

0
Reply
Michael

Ill vouch for Gene...I used a GH3 for a short a few months ago and while the dynamic range is more or less the same out of the box the shadow gradation, sharpness (!!!!), and the fact that it doesn't have a stupid mirror like every other DSLR thus letting you adapt any lens (including PL) imaginable to it just makes the best budget video camera aside from the blackmagic pocket...

oh, and you can use your ipad/iphone as a wireless monitor. like come on. you need a teradek for that shit on other systems

July 17, 2013 at 8:53PM, Edited September 4, 11:21AM

0
Reply
john jeffries

oh and i forgot, it has IN CAMERA OVER/UNDERCRANKING

the only thing i dont like is the laggy touchscreen and stupid menu system

July 17, 2013 at 8:57PM, Edited September 4, 11:21AM

0
Reply
john jeffries

Dollar for dollar the GH is the best camera available, by far. Though I'm not really enamored of it's stills. But the video is so nice to look at. And in the "Zacuto Revenge of the Great Camera Shootout" it, meaning the GH2 Hacked, looked comparable to the ARRI. You can buy something like 35 GH3 bodies with a great lens for the cost of 1 ARRI, that's 35 to 1.

July 17, 2013 at 9:34PM, Edited September 4, 11:21AM

0
Reply
Gene

Not true, obviously. There are many shoots that are well suited for the GH3 and many where it is not. It's a very precise tool not a pocket knife.

July 17, 2013 at 11:06PM, Edited September 4, 11:21AM

0
Reply
Futurusx

I'm not "pro" GH3. I suppose to some it appears I am "pro" GH3.

My sociological/psychological study of the 5D continues.........

July 17, 2013 at 9:39PM, Edited September 4, 11:21AM

0
Reply
Gene

Michael,

I like Fuji cameras over Canon for stills. And I've said that more than once. Why didn't you say I'm "pro" Fuji? Why didn't you say I'm trolling for Fuji? It seems some have GH's on the brain.

July 17, 2013 at 9:43PM, Edited September 4, 11:21AM

0
Reply
Gene

Please don't suggest "best" camera options to users you have never met. Equipment choices are chiefly about what's right for you and your projects. It's about infinite context - some of which really isn't relatable to anybody but the artist.

Your proclaimations do little more than confuse and perplex in situations that are often fraught with anxiety.

The right camera or gear is often the one you actually have. I am sure you know that.

Keep shooting....(hint: not your mouth - use a camera)

July 17, 2013 at 10:55PM, Edited September 4, 11:21AM

3
Reply
Futurusx

I didn't suggest anything to anyone.

I said what I like in cameras. No one asked what I suggest for video and stills. And I didn't offer what I said as a suggestion. I don't know why you got the idea I was from the way I worded what I said.

If you don't like Panasonic for video then that's great for you. If you don't like Fuji for still photography then great for you. I think they are both best, and a sane cost, for both.

Have a nice night. :-)

July 18, 2013 at 12:00AM, Edited September 4, 11:21AM

0
Reply
Gene

You really should use 'the right tool for the job discussion' with the 5D users who use exclusively a 5D for EVERYTHING and say it is the best all around camera you can get. If you suggest there's cameras that are better at this or that than a 5D they will automatically respond, "Yeah but, 5D low light and depth of field" to everything you have to say. They will even say a 5D makes better video than a Red for this and that reason.

Those are the people your counsel will be best for, I think.

July 18, 2013 at 12:13AM, Edited September 4, 11:21AM

1
Reply
Gene

It's all quite impressive but you have to consider that even with the Epic HDRX isn't really usable for alot of situations.

However like I said, still an impressive feat compared to the images you used to get. So I guess they haven't quite yet polished a turd BUT they have rolled it in glitter.

July 17, 2013 at 9:52PM, Edited September 4, 11:21AM

0
Reply

renting ..an arri alexa or arri film camera...in the long wrong will keep you working in the industry with stunning images...and archival footage

July 18, 2013 at 5:43AM, Edited September 4, 11:21AM

3
Reply
DIO

Nah, I'd stick with raw patch, ty.

July 18, 2013 at 11:13AM, Edited September 4, 11:21AM

2
Reply
Natt

without a tripod, the test video has too much motion blur to be an effective testing platform

July 19, 2013 at 9:13AM, Edited September 4, 11:21AM

0
Reply
Dominic

FatRick says, Dam! Luke's wife looks good as a blonde!!

July 21, 2013 at 5:31PM, Edited September 4, 11:21AM

3
Reply
FatRick

The Canon 5D Mk III is nothing short of amazing in it's abilities. Here's a test video I shot showcasing the video capabilities of the camera, https://www.youtube.com/watch?v=1eMbgmiJKXQ

March 19, 2014 at 3:47AM, Edited September 4, 11:56AM

5
Reply

샌디스크 8기가 제품으로 하고 있는데 Lp 인스톨러 초기 화면에서Next버튼이 활성화가 되지 않고 계속 멈춰 있네요...화면에는Please insert a SanDisk ciboatpmle U3 smart drive.라는 메세지뿐... 제 USB 메모리가 잘못된 것일까요?

March 19, 2014 at 9:28AM, Edited September 4, 11:56AM

0
Reply