Magic Lantern enables 14-bit RAW CinemaDNG on 5D

The initial idea was to use HDMI out to bring the 14-bit monochrome signal to an external recorder that can do uncompressed video. For that, they would have to disguise the 14 bits of RAW to make it look like 422 8-bit video. Not easy but not impossible... but then they found that Canon's HDMI-out messes with the image, applying chroma averaging filters and nasty things like that. Since nobody knows exactly which math Canon is applying, this makes it impossible to recover your original 14-bit information from the altered disguised information. So, right now, it's a dead end.

Also, I think I've written this a few times already, but I'll do it again: to give you an idea of how little processing power the ML devs have at their disposal, the initial problem for the 2-second limit was that, if you want to write a standard DNG file, you have to swap byte order (you receive something like 2C-38 and you write 38-2C), and they couldn't do that in real time: frames started to pile up in the buffer, waiting for the byte-swap to end on the previous one, and once the buffer was full, recording had to stop. They solved that by writing the information to a non-standard RAW file without any alterations, then rearranging the data and adding the standard headers and footers with the PC utility.

When you think about that, it is mindblowing what the ML developers are doing here.
 
Samuel H said:
The initial idea was to use HDMI out to bring the 14-bit monochrome signal to an external recorder that can do uncompressed video. For that, they would have to disguise the 14 bits of RAW to make it look like 422 8-bit video. Not easy but not impossible... but then they found that Canon's HDMI-out messes with the image, applying chroma averaging filters and nasty things like that. Since nobody knows exactly which math Canon is applying, this makes it impossible to recover your original 14-bit information from the altered disguised information. So, right now, it's a dead end.

Also, I think I've written this a few times already, but I'll do it again: to give you an idea of how little processing power the ML devs have at their disposal, the initial problem for the 2-second limit was that, if you want to write a standard DNG file, you have to swap byte order (you receive something like 2C-38 and you write 38-2C), and they couldn't do that in real time: frames started to pile up in the buffer, waiting for the byte-swap to end on the previous one, and once the buffer was full, recording had to stop. They solved that by writing the information to a non-standard RAW file without any alterations, then rearranging the data and adding the standard headers and footers with the PC utility.

When you think about that, it is mindblowing what the ML developers are doing here.
Click to expand...

It is really interesting to see how the sensor performs with essentially no preprocessing other than reformatting to DNG standards. The MK3 is really nice. With this approach the limits should only be how fast the data path to the SD card is and how well your media performs.
 
Razz16mm said:
It is really interesting to see how the sensor performs with essentially no preprocessing other than reformatting to DNG standards. The MK3 is really nice. With this approach the limits should only be how fast the data path to the SD card is and how well your media performs.
Click to expand...

CF card, thankfully. Much faster.
 
maarek said:
There is a reason why cheaper cameras will always look "worse" than expensive cameras in motion. The operators. That Alexa flick you posted was stabilized with Gyros. How many DSLR user can do that? Films shot on Alexa are usually thought out a bit more. The shots are planned. Everything is done more properly. Movement is not just pointing the camera somewhere. Why do people have that misconception? If you want filmic motion, you need to move the camera filmicly.
Click to expand...

What I'm talking about is not the way the camera itself is moved. It's the way it renders subject motion in the frame. You could have an Alexa and a DSLR locked down and shooting trees blowing in the wind or someone moving on screen and there will be a difference in the way the cameras render the motion.
 
RAW is a great tool, I only shoot RAW for my still work, but it might be pushing the limits of this camera. I would be completely happy with a DR and detail increase using the motion jpeg container for an easy workflow and longer recording times. But I will be thankful for any improvements. Great to see the excitement. Very cool.
 
^ That seems to be the case right now, since the file system won't allow files of more than 4GB. But I would expect ML to implement file spawn for continuous recording once everything else is sorted out.
 
Yeah it's temporary, RAW is barely out of the proof of concept stage, we've only been testing for like 4 days.
 
I've been able to record a lot longer than a minute using a 400x Transcend Card on my 5D3... There have been uploads already with Audio,... I've been able to do shoot in the RAW Container on my Mark 3 and my Mark 2... It is getting better by the minute...
I dropped $4000.00 yesterday on a Mark 3, I was so wanting to buy the Black Magic Cinema Camera 4K,... I mean really struggling,... Who knows when it will ship? I just can't see tying up that kind of money,... I was even considering the FS700, F3, and of course the C300 and the Scarlet... I may still buy one or all,(I've flipped a lot of gear in the Market Place)... I have to admit, for what I do (Documentary, Local Television and Web Commercials and Wedding Films) the 5D Markiii will with Magic Lantern will get a lot of use.
I will Use the Following because of the Magic Lantern - Focus Peak, Zebras, Scopes, Continuous Recording, Audio Functionality (The ability to gain on Channel Higher than the other), FPS Frames Per Second Override, HDR Video, Live View Color Displays for shooting with LOG Style Picture Styles, Intervolometer, WAV Audio Recorder, Did I mention Focus Peak? Good Grief that is a life saver...
I am truly going to love the RAW Feature on Magic Lantern, its just another tool to use in my bag of tricks... Way to go Alex and 1%!
 
Samuel H said:
^ That seems to be the case right now, since the file system won't allow files of more than 4GB. But I would expect ML to implement file spawn for continuous recording once everything else is sorted out.
Click to expand...

I highly doubt that....that would double the bandwidth requirements for the brief moment where the files switch over. I suspect you'll deal with a break in the recording, just like ML does for the camera's internal recording mechanism.
 
Gary Huff said:
I highly doubt that....that would double the bandwidth requirements for the brief moment where the files switch over. I suspect you'll deal with a break in the recording, just like ML does for the camera's internal recording mechanism.
Click to expand...

The 4GB file limit is a constraint the Magic Lantern people put on it while its in experimental stage. They will most likely remove it at some point once they understand the RAW stuff a little more.

Above is what I gathered.
 
Another test that needs to be done when/if they get full recording at the highest rate to go past the 4GB limit: Continuous recording over heat test...not from the processing, but from the card bay. Not sure anyone is that excited to do that test on a $3500 camera though..lol.
 
yoclay said:
Concerning the Alexa. It has something most video cameras don't. A mechanical shutter. (And an optical viewfinder)
This means that the way the image is exposed is actually much closer to the way film is shot.
So in a sense it is neither rolling shutter or global shutter, but a bit of both.
This is a big part of why it does not have the same look as a conventional video camera.
Click to expand...

Only some models of Alexa do, but all cameras have a signature way of interpreting motion blur that comes from how they record and interpret the fact that different colors of light are hitting the sensor for varying amounts of time during the interval when the shutter, electronic or mechanical, is open- a debayered image from any CMOS camera is an interpolated guess on what the sensor "saw", never a literal translation. On Red One for example, the motion blur looked different (and better) after a firmware upgrade a few years ago.
 
pulpfiction007 said:
Another test that needs to be done when/if they get full recording at the highest rate to go past the 4GB limit: Continuous recording over heat test...not from the processing, but from the card bay. Not sure anyone is that excited to do that test on a $3500 camera though..lol.
Click to expand...

It's not just the card bay, but the digics themselves converting sensor data to raw recorded information. ML RAW re-purposes the digic and maxes it out for prolonged periods never intended by Canon. Any circuit can overheat. MY Cuda cards need very high fan speeds to not crash when playing long Resolve timelines. There is also the matter of the card itself running at peak speeds for extended periods. I don't mean to be a pessimist, but there is a lot that could go wrong on long takes here.
 
So how long does a single 4GB file last when recording in 5D Mk3 1080 HD RAW ?

From what I've read it sounds like 4GB buys you roughly 30 seconds of recording time.
 
Robert Ruffo said:
It's not just the card bay, but the digics themselves converting sensor data to raw recorded information. ML RAW re-purposes the digic and maxes it out for prolonged periods never intended by Canon. Any circuit can overheat. MY Cuda cards need very high fan speeds to not crash when playing long Resolve timelines. There is also the matter of the card itself running at peak speeds for extended periods. I don't mean to be a pessimist, but there is a lot that could go wrong on long takes here.
Click to expand...

Dealing with raw data is way less processor intensive than real-time encoding H264 though. Data bandwidth on the other hand...
 
You must log in or register to reply here.
Back
Top