Magic Lantern enables 14-bit RAW CinemaDNG on 5D

Gotta say, so far the H.264 looks better on this camera than the raw.

What did you use to process the raw footage?
 
Indeed, those 5D2 artifacts are horrible... but they clear out pretty well with the filter. And in any case, the beauty of shooting RAW is that you have options: the fact that the H.264 clip was shot with the same information input proves that you can reduce the artifacts just by using a different debayering method (one that will kill a lot of detail, I'm afraid).
The Adobe Camera RAW plugin usually has some sharpness turned on by default. Turning it off might help. And maybe it can go negative?
 
yes I did, thanks
still playing with them, I'll post some thoughts soon

edit:
ok, here we go...

Ted sent me a couple of SRAW2 images taken in stills mode, and also a couple of LV-RAW images from the Magic Lantern hack. I converted the CR2 files to DNG, then processed them all with the same settings (trying to extend DR as much as possible, and with sharpness at 0). Then I downsampled the SRAW2 images using bicubic, and cropped everything to 16:9 (well, actually, just close to that, as I wanted to bring the FoV somewhat closer to each other).

This is without the mosaic VAF:
http://www.similaar.com/aarchive/5d2_lvraw.jpg
http://www.similaar.com/aarchive/5d2_sraw2_bicubic.jpg
And this is with the mosaic VAF:
http://www.similaar.com/aarchive/5d2_lvraw_mosaic.jpg
http://www.similaar.com/aarchive/5d2_sraw2_bicubic_mosaic.jpg

Comments:
* Color is basically identical. Anybody complaining that the RAW videos have awful colors is complainign about the grading, not the camera's abilities (any look you can achieve on a RAW still, you can achieve it just the same with this LV-RAW thing).
* No idea about DR or noise in the shadows, a res chart is not the place to look at those issues.
* There's some aliasing and moire in the LV-RAW images, but not nearly as much as shown in previous samples; still an issue, but not any worse than with the regular H.264 video from this camera (reminder: it's a 5D2).
* The mosaic VAF gets mostly rid of those aliasing/moire issues, but there's a slight effect on sharpness too.
* In the end, if you compare the LV-RAW image with the filter vs the SRAW2 image without the filter (links 3 and 2), well, it's clear the former one is softer. But it's not too far away, is it? I would say one is close to 800 lines, the other close to 900 lines.

Summary: not bad, but I think the 5D3 will be better (if this actually ever truly works)
 
Last edited:
Ted, how did you get 51 frames? Is this through burst ? Higher class card?

Joe, yes through burst end trigger mode. For now its consistent at 51 using only a 133x cf card.

added: - also set the canon menu NOT to shoot raw and set JPG at lowest setting. If I set to shoot raw, I will only get 39 frames.
 
Last edited:
mmm... if the number of frames you can record depends on the stills uality option you have selected, my guess is that processing power for this is what is holding down the continuous RAW video option.
What I don't understand is why the image has to be processed to still in order to record the RAW DNG: the live-view buffer is there all the time, whether you're recording or not! (unlike on Sony cameras , for example).
It seems like if ML can turn off that processing, we will have continuous RAW video (at least on very fast cards).
 
My article about this just went live today:

http://www.redsharknews.com/technology/item/684-raw-video-from-the-canon-eos-5d-mk-iii

I see a lot of very sensible stuff being said in this thread. Everybody seems to go a bit mad about RAW but it's not actually the right tool for every situation.

In this case, with only about 15 seconds of 24fps this is probably mostly going to only be genuinely useful in a practical way for certain kinds of music video's I suspect.

Having said that I think this is going to be great for people who already have an EOS camera to try out RAW and see if the workflow works for them before taking the leap to trying a camera more generally RAW based.

They may find it's not for them and save themselves some bother for instance. :)

Freya
 
What I don't understand is why the image has to be processed to still in order to record the RAW DNG: the live-view buffer is there all the time, whether you're recording or not! (unlike on Sony cameras , for example).
It seems like if ML can turn off that processing, we will have continuous RAW video (at least on very fast cards).

Well the live view is debayered video (and scaled AFAIK too!).

Also as I understand it the Raw video they have discovered is actually already in RAW DNG format (not cinema DNG). I think the biggest problem with RAW is going to be the speed at which the Compact flash cards can be written to. Following that the speed the processor can go at. See my article in the post above for a better explanation.

Freya
 
Well the live view is debayered video (and scaled AFAIK too!).

Also as I understand it the Raw video they have discovered is actually already in RAW DNG format (not cinema DNG). I think the biggest problem with RAW is going to be the speed at which the Compact flash cards can be written to. Following that the speed the processor can go at. See my article in the post above for a better explanation.

Freya

cinemaDNG raw is the same as DNG raw except for metadata to attach and sync audio files and define clip sequences for motion playback. It groups each set of sequential DNG frames that make up a clip along with audio and metatdata and saves them together in a separate file folder. There is no real difference in the data that defines the individual frames.

They would have to be picking up the sub-sampled raw sensor data before it is deBayered and encoded and wrapping it in DNG format.
 
Last edited:
cinemaDNG raw is the same as DNG raw except for metadata to attach and sync audio files and define clip sequences for motion playback. It groups each set of sequential DNG frames that make up a clip along with audio and metatdata and saves them together in a separate file folder. There is no real difference in the data that defines the individual frames.

They would have to be picking up the sub-sampled raw sensor data before it is deBayered and encoded and wrapping it in DNG format.

Correct: cinemaDNG, as used in the BMC and by ML, is just a series of DNG stills with a companion audio file. The spec also supports a single-file format (IIRC, it uses MXF), but nobody's using that AFAIK.

Plus: they didn't catch a DNG that was floating somewhere, they found a dev flag that enabled the use of a buffer which contained what looked like random useless data, then, through trial and error, made sense of what was there, re-arranged it, and stored it to disk with a DNG header so the file can be read by standard software.

Finally, to elaborate on what I don't get: with that flag on, the camera is putting all that info on the buffer, and at the same time displaying the live-view correctly debayered on real-time. When you start recording the RAW stream, why does extra stuff have to be done? (i.e. whatever extra work is usually done to save RAW or JPG stills, is done now too, for some reason)
 
12 seconds of RAW recording! The OP says he's getting continuous 1928x850 24p RAW recording on a 1000x card!

 
Last edited:
This is the biggest thing since the 5D MKII came out. "Game changer" :cheesy:.

After I started shooting 2 second RAW clips I decided I could never go back to H.264.
 
Last edited:
Has the poster clarified how long he can shoot...if it truly is continuous, why aren't people more excited...lol. That would be a very big deal. Lots of people might jump ship from BMC and all of their issues with shipments. I would be happy to say I was wrong about this working at all (for meaningful segments of time that is)
 
It only just happened, his is the first footage of more than 2 seconds posted and the build he is using isn't readily available yet. At this point there isn't even an app with a GUI that can read the RAW files, you have to use a Linux command line utility but it shouldn't be long before someone writes mac and windows utilities.

I wouldn't break out the champagne just yet, we don't know how reliable it's gonna be and whether it will cook the digicV.

The 1928x850 lv_rec DNGs are 3.08Mb per frame down from 4.7Mb in silent pic burst mode.
 
The OP wrote-

​
"I did a test to see what would happen if I filled the memory card. It records the first file fine. It keeps recording until the memory card is full, but I am having strange results after the first file. I just sent a1ex a message about it."
 
Back
Top