Magic Lantern enables 14-bit RAW CinemaDNG on 5D

lawriejaffa said:
Hey gents I was wondering - what kinda stable recording times are you finding possible on your tests at RAW 1080 (or likely possible with a raw 1080 2:35 crop?)

A bit of naysaying that the feature will be of any practical use?
Click to expand...

I can record uncropped 1080p on the MK3 on 1066x card till it fills up every single time, multiple files are fine too. There's a lot of work still to be down before an alpha is released but 1080p recording is completely stable in my experience.

On the topic of comparison tests. I haven't done any since the day I started shooting raw, there is no comparison.

On the image quirk topic. There have been at least a couple of builds that have been buggy which is to be expected considering there are builds coming out daily. This development is still in the pre-alpha test stage so a lot of different ideas are being tried like 12 bit for instance to improve on what is already a great thing. Most of the builds have been rock solid with no image artifacts or freezes or dropped frames. I only realised the other day that the exposure flickering that's visible in all my footage is actually an After Effects bug and not ML.
 
Last edited:
For those who are interested, I decided to jump into the party and throw the latest build onto the ole 5D2. I am shooting with SanDisk Extreme 60mb/s cards. The highest res I can get continuous shooting with is 1720x720 which is almost 2.40:1. Right away I noticed that moire and aliasing is worse than straight H.264. That is really the deal breaker at the moment. I used a lightroom workflow, exported tiff 8 bit and imported to premiere for grading. I decided to try some film convert on it to see if some grain would help cover up some of the issues...it did not. Anyway. This is the super quick and super dirty test. Also the weird motion is handheld with IS on.

Forgot to mention that there is 0 sharpening of any kind added.

 
Andrius Simutis said:
I believe the bottleneck for SD cards is actually before the card at the interface, so no matter how fast the card is it won't make any difference. It's kinda like having a super fast RAID connected by USB2.
Click to expand...

You're right. After I posted I did some research and found that Canon didn't make the SD slot to UHS-I spec. Oh well. In addition to RAW video that will probably be the 5Dmk4 killer feature.

The UHS-II spec is crazy. Up to 300MB/s and up to 2TB cards. Of course they are not available... yet...
 
Last edited:
Steve Kahn said:
You're right. After I posted I did some research and found that Canon didn't make the SD slot to UHS-I spec. Oh well. In addition to RAW video that will probably be the 5Dmk4 killer feature.

The UHS-II spec is crazy. Up to 300MB/s and up to 2TB cards. Of course they are not available... yet...
Click to expand...

That is like CFAST, the SATA version of Compact Flash, it can do 600MBps currently, but it isn't supported by very many devices yet.
 
Rolling shutter test 5d2 shot with raw video vs h264

Rolling shutter test 5d2 shot with raw video vs h264

For those curious to know about the rolling shutter "bendy thingy" in raw dng mode, I made this test with 2 cameras camparing raw video vs h264.

Since I only have one 5D2 and a 7D, I shot them first both at H264 to create a baseline behavior.
If both have similar behavior in h264, then I can use the 7D to do a side by side with the 5D recording raw while the 7D at h264.

You can draw your own conclusions.

 
(damn these two raw threads!...)

All, someone else did a rolling shutter test like this a while back (think it was for a GH2/GHh3 test) and I mentioned that mounting a camera on a hot shoe and then whipping it around was bound to add extra movement for the top camera. He was irritated with me at first but did reverse the cameras, and indeed, the top camera always had the worst rolling shutter. The hot shoe is just not solid enough a mount for this application.
 
Cosimo Bullo said:
(damn these two raw threads!...)

All, someone else did a rolling shutter test like this a while back (think it was for a GH2/GHh3 test) and I mentioned that mounting a camera on a hot shoe and then whipping it around was bound to add extra movement for the top camera. He was irritated with me at first but did reverse the cameras, and indeed, the top camera always had the worst rolling shutter. The hot shoe is just not solid enough a mount for this application.
Click to expand...

Also added this to the other thread :)

The result I was after is not to see which camera was better but rather to have a basis to see if RS was any worse for just one camera (the 5D) between two formats, Raw DNG and H264.

The result I got for the first test, which was H264 on both, was what I used as basis to see in the second, which is raw and H264 if there is a difference between raw and h264 for the 5D alone.

That was my objective. Not as weather one cam had better RS effect than the other.
 
Cosimo Bullo said:
@Squig - Looks like some are reporting the vertical band/strips problem is gone with the latest build. Can you confirm?

http://www.magiclantern.fm/forum/index.php?topic=5490.0
Click to expand...

Yeah it's gone but there's a random pixel issue that appeared after the build on the 17th. Devs are looking into it. There's a few bug free builds so it shouldn't be too hard to sort out. I did a 12min external recorder sound sync test today and the sync was perfect.
 
Ted Ramasola said:
Also added this to the other thread :)

The result I was after is not to see which camera was better but rather to have a basis to see if RS was any worse for just one camera (the 5D) between two formats, Raw DNG and H264.

The result I got for the first test, which was H264 on both, was what I used as basis to see in the second, which is raw and H264 if there is a difference between raw and h264 for the 5D alone.

That was my objective. Not as weather one cam had better RS effect than the other.
Click to expand...

Still, that method has way too much room for error... Do you want to put a number on the RS issue? Still not perfect, but better, I think.
http://www.dvxuser.com/V6/showthrea...g-rolling-shutter-put-a-number-on-this-issue!
(if you send me something, please do longer pans, on these short kind it's nearly impossible to get constant speed, the camera is always accelerating or decelerating)
(if you want to send 5D-H264, 5D-RAW, and 7D, I'll measure them all)

edit: note that this could very well be all pointless, since I believe sensor readout speed does not depend on what you do with that information somewhere down the pipeline (store it as h264 or raw), but anyway, if you're still wondering if there's a difference, it's not that difficult to find out for sure
 
Last edited:
squig said:
I haven't bothered doing the RS test cause I'm 99% sure it's exactly the same.
Click to expand...

after doing that simple test, RS is a closed issue for me now as well. :) -visually it is similar in raw vs h264. If i saw a huge discrepancy, I would have sent an alarm to the ML devs.
 
Just shot my best footage yet. Here's a teaser.

TPK6r2R.jpg
 
As terrific as RAW may be, I'll be sticking with compressed H.264 for a while. For what I do RAW simply isn't worth the trouble most of the time. Sure, for the occasional wide establishing shot, yeah...But there's no way I would shoot an entire film with raw. I think many don't realise just how painful dealing with RAW is. Don't get me wrong, I love the 5d3 and i'm awed by what magic lantern has done, but it just isn't practical for most stuff. For narratives, yeah...but for everything else....not really. In my opinion, at least. Your average viewer definitely won't notice the difference.
 
It is still early in the game. As the right tools emerge and workflows get established it will become less painfull to shoot in RAW for video. I do believe everybody will be able to tell the difference if the RAWs are processed and shot to maximize DR. Your point is true, nobody is going to shoot a full soccer game with RAW but over time folks will learn when to best fit it in their workflow.

I only shoot RAW for stills, so the workflow is very familiar to me. I shot a week long timelapse with RAW stills, processed them in one swoop Lightroom, exported jpegs, imported the jpegs as a clip in Edius and then edited as a video file. Yes it was a bit long but I would not say painfull because the work as background automated and the image quality was awesome.
 
I can record uncropped 1080p on the MK3 on 1066x card till it fills up every single time, multiple files are fine too. There's a lot of work still to be down before an alpha is released but 1080p recording is completely stable in my experience.

On the topic of comparison tests. I haven't done any since the day I started shooting raw, there is no comparison.

On the image quirk topic. There have been at least a couple of builds that have been buggy which is to be expected considering there are builds coming out daily. This development is still in the pre-alpha test stage so a lot of different ideas are being tried like 12 bit for instance to improve on what is already a great thing. Most of the builds have been rock solid with no image artifacts or freezes or dropped frames. I only realised the other day that the exposure flickering that's visible in all my footage is actually an After Effects bug and not ML.
Click to expand...

Hey Squig thanks for taking the time to reply - appreciate getting first hand info from those experimenting with the pre-alphas. It's great to know that has been your experience. I shoot low budget features, and (have a decent rig for editing in post/grading etc) so being able to shoot RAW for this purpose on a 5d mk3 is a very very attractive prospect!

I think those bemoaning (whether they really need RAW) are missing the point... if it's working effectively then it's simply an essential addition for many filmmakers and a brilliant option. For many it's the entire reason behind their purchase of camera formats such as BMC. So it's exciting to see where this may take us!
 
You must log in or register to reply here.
Back
Top