September 02, 2014, 06:23:23 AM

Author Topic: ML apparently discovers hook to 'raw' liveview feed  (Read 3090 times)

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
ML apparently discovers hook to 'raw' liveview feed
« on: April 26, 2013, 04:23:37 PM »
http://www.eoshd.com/content/10221/magic-lantern-discover-2k-raw-dng-function-on-5d-mark-ii-and-iii
http://www.eoshd.com/comments/topic/2604-magic-lantern-discover-2k-raw-dng-stream-in-live-view-on-the-5d-mark-iii/

Now I don;t know if they will be able to get video out of this but as they say here it at least brings up some curious questions:

"Image quality is very good – pin sharp 2K. Why the hell is this feature not implemented if the cameras are capable of it?? Also if the sensor can provide 2K DNG in live view, why is the video mode so soft and challenged for dynamic range? I find this very puzzling."

Bizarre indeed (if what they say is true and there are not catches, perhaps it is pin sharp but aliased, at least on the 5D2 or maybe on both???)!

If it can make pin sharp full 2K frames at 24fps rate on the 5D3 and even 5D2 then why does their video engine end up putting out on the 5D2 hugely aliased, trace soft HD and the 5D3 output a soft but clean HD and why in both cases is it only 8bits and not at leat 10bits and why is the DR so extremely limited when after so much downscaling it should be much better you'd think. Is Canon marketing crippling cameras even FAR more than we ever imagined?? Is there some minor technical detail (could well be but if so might not some relatively inexpensive extra few parts not have fixed it)?

Maybe they are leaving out some crucial bit of info since it seems hard to believe, especially regarding the 5D2, that Canon would hide the ability to get sharp and aliasing free video.
« Last Edit: April 27, 2013, 01:15:12 AM by LetTheRightLensIn »

canon rumors FORUM

ML apparently discovers hook to 'raw' liveview feed
« on: April 26, 2013, 04:23:37 PM »

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
Again keep in mind that actually getting 14bit 1080p uncompressed video still seems very unlikely.

But if you look in this thread http://www.fredmiranda.com/forum/topic/1208244/0#11512420 someone posted a frame captured from the liveview RAW 24fps feed and it looks so much crisper than the 5D3 video. It seems to imply that they actually are getting something amazing off of the sensor at 1080P. So why does, as some say, the video look like a touch of gaussian blur was applied to it? Why don't we get that perfect crispness that the camera seems to be able to capture at 24fps? (And why didn't they put 10bit 4:2:2 chip in the 5D3 after they say how 5D2 video took off?)

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
https://www.dropbox.com/s/5c2vah29wis0wjk/52090007.DNG

How does that signal that the hardware apparently can deliver then turn into twice the mush?

HurtinMinorKey

  • 7D
  • *****
  • Posts: 413
    • View Profile
    • carolineculler.com
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #3 on: April 27, 2013, 03:31:37 PM »
Apparently this Raw Feed gets written to a buffer, but the buffer is 512MB, and can only hold a few seconds.  Keep in mind, that SD and CF write speeds aren't fast enough to write 14bit raw at this resolution. So at the end of the day, I'm guessing that the most we'd get from this is a Raw burst mode similar to the Nikon 1, v1 (or v2).

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #4 on: April 27, 2013, 04:52:51 PM »
Apparently this Raw Feed gets written to a buffer, but the buffer is 512MB, and can only hold a few seconds.  Keep in mind, that SD and CF write speeds aren't fast enough to write 14bit raw at this resolution. So at the end of the day, I'm guessing that the most we'd get from this is a Raw burst mode similar to the Nikon 1, v1 (or v2).

It would be nice if there were a way to clip it to 1920x1080 and feed it to the h.264 compression engine though without doing whatever horror it is that apparently gets done sometime between the 'raw' buffer and the h.264 engine. Because that 'raw' buffer stuff sure seems a lot crisper with better micro-contrast for fine details and less waxy.

I'd think it would be possible if you had full documentation for everything and deep control. As much as ML has figured out there is a lot of stuff they can't touch. Maybe they have enough to do this, maybe not though.

I wonder if it could even be fed at 10bit.

If the 'raw' was clipped to 1920x1080 and 8 bits that might just make the file small enough to dump into the 512MB buffer and flush on to the CD card quickly enough to maintain 24fps. I'm not what the detail of this 'raw' feed are and what processor can work on it and whether the ones they have access to could do that or do that quickly enough. I'm pretty sure they have no control over programming the DIGIC image processor.


LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #5 on: April 27, 2013, 04:56:48 PM »
It also makes you wonder if they may not have been able to make a 5D3 or at least a 5D3C with maybe a 1536MB buffer which maybe could hold enough for it to flush to the CF card and taken the cinema world by storm as with the 5D2 and had them flying off the shelf even faster than the 5D2. Or tweaked whatever part is the hold up. They tend to be in sit and react and sit and milk mode instead of lets go full steam ahead and revolutionize everything and conquer and dominate new markets. Who knows.



BrettS

  • PowerShot G1 X II
  • ***
  • Posts: 72
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #6 on: April 27, 2013, 05:20:18 PM »
They tend to be in sit and react and sit and milk mode instead of lets go full steam ahead and revolutionize everything and conquer and dominate new markets.
It does seem that way, yeah.

canon rumors FORUM

Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #6 on: April 27, 2013, 05:20:18 PM »

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #7 on: April 28, 2013, 03:15:19 AM »
Quote
sqdstf wrote:
http://youtu.be/YOLDDrfpFO8

He seems to be too optimistic though. Sounds like 12-14fps may be the max. (unless they can find a way to feed  a 1920x1080 cropped version of those DNGs into the compressor engine and the engine doesn't clip to 8bits even if it did that at least it might maintain much better details)

But man I just don't get why Canon seemingly left so much on the table. Sure uncompressed is too much for the HW to handle but why in the world could they not have written out a compressed file with original DR left intact and not utterly massacred and with a few more bits? And what on earth are they doing to reduce the beautiful crisp 1920x1080 the fancy sensor the engineers developed for them and then turning it to utter mush! It's not the compression engine because using the HDMI out with ProRes HQ doesn't really seem to bring back any extra detail all (certainly not static detail). It just seems like they left a ton of ability on the table. They would've had an utter revolution if they had not mushed down the resolution that it is actually capturing and written it out with more bits and less DR compression done and put out a really crips compressed file with lots of DR room.

Maybe this DNG thing is not what they are able to use to drive the compression engine and some crappier source must be used?? I don't get it otherwise.

Axilrod

  • 1D X
  • *******
  • Posts: 1373
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #8 on: April 28, 2013, 03:27:21 PM »
Quote
sqdstf wrote:
http://youtu.be/YOLDDrfpFO8

He seems to be too optimistic though. Sounds like 12-14fps may be the max. (unless they can find a way to feed  a 1920x1080 cropped version of those DNGs into the compressor engine and the engine doesn't clip to 8bits even if it did that at least it might maintain much better details)

But man I just don't get why Canon seemingly left so much on the table. Sure uncompressed is too much for the HW to handle but why in the world could they not have written out a compressed file with original DR left intact and not utterly massacred and with a few more bits? And what on earth are they doing to reduce the beautiful crisp 1920x1080 the fancy sensor the engineers developed for them and then turning it to utter mush! It's not the compression engine because using the HDMI out with ProRes HQ doesn't really seem to bring back any extra detail all (certainly not static detail). It just seems like they left a ton of ability on the table. They would've had an utter revolution if they had not mushed down the resolution that it is actually capturing and written it out with more bits and less DR compression done and put out a really crips compressed file with lots of DR room.

Maybe this DNG thing is not what they are able to use to drive the compression engine and some crappier source must be used?? I don't get it otherwise.

It has something to do with the buffer, but from what I've read most seem confident they will be able to get it to work.
5DIII/5DII/Bunch of L's and ZE's, currently rearranging.

Roger Jones

  • PowerShot G1 X II
  • ***
  • Posts: 55
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #9 on: April 28, 2013, 07:25:21 PM »
At 5mb per frame 24p footage would be ~ 120MB per second or ~8GB per minute. Cropping the frame rather than just writing the raw dump is a time consuming process. Its faster to just write the data rather than write a cropped window of it. Judging by the image size this kinda looks like a 3 pixel bin of the sensor data. New black magic cameras record compressed raw. Not clear if digic 5 has enough power to do this.

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #10 on: April 29, 2013, 01:23:51 PM »
At 5mb per frame 24p footage would be ~ 120MB per second or ~8GB per minute. Cropping the frame rather than just writing the raw dump is a time consuming process. Its faster to just write the data rather than write a cropped window of it. Judging by the image size this kinda looks like a 3 pixel bin of the sensor data. New black magic cameras record compressed raw. Not clear if digic 5 has enough power to do this.

But there probably isn't a way to make the just dump the whole thing to the buffer to work for more than a second or two. But maybe you could clip the frames and send them over HDMI or feed them to the h.264 compression engine to get around those issues in part, in the latter case obviously no longer uncompressed and likely not 14bit but at least it might be crisper and perhaps it could even be 10bit and with more DR.

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3533
    • View Profile
Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #11 on: April 29, 2013, 09:02:42 PM »
I wonder if the new firmware doesn't actually make things a touch sharper after all, maybe comparing HDMI vs new internal is the same but maybe new internal is sharper than old? Fine edges seem to get mosquito noise sort of, maybe part of making it sharper. I need to shoot with the new, rollback and shoot with the old and compare. Maybe they did sharpen it up a bit? It is a different look than the 'raw' though, less fine micrcontrast still and yet maybe more aliasing than old video or the 'raw' video buffer?

EDIT: the artifacts, fine detail shimmer and such seem to be something going wrong in Premiere Pro (h.264, High, 5.1, 40Mbs CBR; USM/sharpen/colorista tools) since both the original Ninja recorded files and ones recorded by the 5D3 with new firmware DO NOT have that weird stuff going on. So whatever that is it is definitely not the fault of the 5D3 and/or new firmware.

EDIT: EDIT: well other than in one file where there is some weird black band that pops on and off across a few letters of white text in a weird way on one file, that is on the original
MORE EDITS: actually that weird stuff is in all the files, slight bumps to cam and almost like a bit of 5D2 line skip sorta aliasing jitter, just without sharpenin git wasn't as easy to spot
« Last Edit: April 30, 2013, 03:42:31 AM by LetTheRightLensIn »

canon rumors FORUM

Re: ML apparently discovers hook to 'raw' liveview feed
« Reply #11 on: April 29, 2013, 09:02:42 PM »