Author Topic: Danne's crop_rec_4k & ISO experiments  (Read 63665 times)

ozcancelik

  • Freshman
  • **
  • Posts: 90
Re: Danne's crop_rec_4k & ISO experiments
« Reply #650 on: September 14, 2020, 11:12:55 PM »
Crash has become much less. Doesn't crashed when changing global draw. It "sometimes" crash when changing FPS. I was able to record 24 frames 3264x1366 continuously. (8 minutes). There is overlay pattern when recording but it does not prevent shooting. Also image becomes greenish.

Also it would be nice if the size that is seen on the screen and the size taken is the same.

Got word from has_abandonment_issues about 3.3k preset working with this version. Missed uploading yesterday:
New builds:
https://www.magiclantern.fm/forum/index.php?topic=23041.msg216899#msg216899


Danne

  • Contributor
  • Hero Member
  • *****
  • Posts: 6965
Re: Danne's crop_rec_4k & ISO experiments
« Reply #651 on: September 14, 2020, 11:39:06 PM »
Try and find the answers to why fps makes your camera crash, also why changing global draw is affecting your recordings etc. With time and testing you´ll find your answers.

In general. All you are describing is already obvious to most users. If it was possible to change it it would have been changed already.

For real time preview check out theBilalFakhouri work.

Bender@arsch

  • Freshman
  • **
  • Posts: 84
Re: Danne's crop_rec_4k & ISO experiments
« Reply #652 on: September 15, 2020, 12:16:45 PM »
Got word from has_abandonment_issues about 3.3k preset working with this version. Missed uploading yesterday:
New builds:
https://www.magiclantern.fm/forum/index.php?topic=23041.msg216899#msg216899

Realy nice! All works!
I can continuous shoot at 3.3K (16:9) in 12bit and 3.5K (2.07:1) in 10bit
 - >real world test

Interestingly the writespeed indicator is up to 137,5 mb/s, but I don't double check it on my computer.

Dmytro_ua

  • Member
  • ***
  • Posts: 202
  • Kyiv, Ukraine
Re: Danne's crop_rec_4k & ISO experiments
« Reply #653 on: September 15, 2020, 12:37:15 PM »
What about noise amount? I've just made a few quick shots comparing to anamorphic mode and the noise in shadows looks more noticeable than in anamorphic. Though didn't have enough time to check it properly yet..
5d3 1.1.3 exp | Canon 16-35 4.0L | Canon 50 1.4 | Canon 100mm 2.8 macro
Ronin-S

Bender@arsch

  • Freshman
  • **
  • Posts: 84
Re: Danne's crop_rec_4k & ISO experiments
« Reply #654 on: September 15, 2020, 12:40:23 PM »
Ok, i double check it:

I search a good spot (bright and flat spot, without dark areas) and show in ML Menu under Raw -> 14Bit writespeed preview say 208,5mb/s or so

Here the Result:
3.5K 3584x1730 23,976fps 10bit
CF: 81,9
SD: 55,6
=137,5MB/s
-> 8424 (cf 5029, sd 3394), 1469 frames =~61sec
8424/61 =138mb/s !!!

This is a new record!!!

p.s. the indicator was orange and i stoped manuel;)

Grognard

  • New to the forum
  • *
  • Posts: 28
Re: Danne's crop_rec_4k & ISO experiments
« Reply #655 on: September 15, 2020, 01:37:12 PM »
UHD 2.39 is "only" 6.170 880 pixels (3840x1607) less than in this case : 6.200 320 pixels (3584x1730)

Bender@arsch

  • Freshman
  • **
  • Posts: 84
Re: Danne's crop_rec_4k & ISO experiments
« Reply #656 on: September 15, 2020, 01:47:34 PM »
UHD 2.39 is "only" 6.170 880 pixels (3840x1607) less than in this case : 6.200 320 pixels (3584x1730)

I don't know what you're trying to say, but the important thing is the increased writespeed. And the latest build can only 3840x1536 (continuous recording too).

-> we got never more than 131mb/s

has_abandonment_issu

  • New to the forum
  • *
  • Posts: 28
Re: Danne's crop_rec_4k & ISO experiments
« Reply #657 on: September 15, 2020, 02:01:01 PM »
I don't know what you're trying to say, but the important thing is the increased writespeed. And the latest build can only 3840x1536 (continuous recording too).

-> we got never more than 131mb/s

I didn’t want to overstate anything but it seemed to me also that the write speeds were faster than I expected them to be in 3.5k. Thanks for actually checking!

I think there is something about the 3.5k preset that the 5d just handles extremely well. Less system load or something.

Danne

  • Contributor
  • Hero Member
  • *****
  • Posts: 6965
Re: Danne's crop_rec_4k & ISO experiments
« Reply #658 on: September 15, 2020, 02:52:34 PM »
Ok, i double check it:

I search a good spot (bright and flat spot, without dark areas) and show in ML Menu under Raw -> 14Bit writespeed preview say 208,5mb/s or so

Here the Result:
3.5K 3584x1730 23,976fps 10bit
CF: 81,9
SD: 55,6
=137,5MB/s
-> 8424 (cf 5029, sd 3394), 1469 frames =~61sec
8424/61 =138mb/s !!!

This is a new record!!!

p.s. the indicator was orange and i stoped manuel;)
Those are nice numbers. I am pretty sure it´s because of being in x5 zoom.

IDA_ML

  • Hero Member
  • *****
  • Posts: 885
Re: Danne's crop_rec_4k & ISO experiments
« Reply #659 on: September 15, 2020, 06:42:10 PM »
I can also confirm that the 3,5k and 3,3k modes work really well on the 5D3.  I get similar results as Bender@arsch, just 2-3 MB/s less since I only have the old 95 MB/s SD-card.  And I did not have to overclock it. I also get continuous recording at 3200x1800/10bitLL/24fps (16:9) and more than 400 frames recorded at 12bitLL and 0.0 exposure without a single corrupt frame.  Thank you so much, Bilal and Danne for this amazing work and for bringing the 5D3 that far.  I still cannot believe that the 5D3 can do this.

The only thing that I am missing in this latest build is the slow choppy preview of the flv anamophic mode which is a full-sensor readout mode.  I use it a lot at 15 fps and it really works very well for wide angle landscape videography since it allows slow camera panning and provides very smooth results at 1/30-th of a sec. shutter speed.  Unfortunately, flv preview becomes skewed when recording is started and I don't how to fix it.   So, I was wondering if the marvelous 5x real time preview could be combined with the slow flv preview during recording so that both could be toggled with the INFO button.  Also, I was wondering if the frame rate could be raised by another 2-3 fps in that flv anamorphic mode.  Right now, write speed is about 75 MB/s (green indicator) at 15 fps, so there seems to be a lot of overhead for higher fps available.  The EOS-M does conveniently 16.688 fps in its flv anamorphic mode and is very stable at that frame rate, no corrupt frames either.  In my experience, 17-18 fps frame rate seems to be a sweet spot for optical flow interpolation to 24 fps in Resolve.

70MM13

  • Senior
  • ****
  • Posts: 484
Re: Danne's crop_rec_4k & ISO experiments
« Reply #660 on: September 15, 2020, 07:05:27 PM »
resolve does 18fps timelines so that would be a great choice!  i use it sometimes instead of 23.976 and it is really nice.

has_abandonment_issu

  • New to the forum
  • *
  • Posts: 28
Re: Danne's crop_rec_4k & ISO experiments
« Reply #661 on: September 15, 2020, 09:11:52 PM »
When recording you can press the magic lantern menu button (trashcan) and see a current recording speed as well as a max recording speed. How accurate is that using the 3.5k preset?

Edit: Tested it myself, what I'm talking about is if you move your selector over the RAW module, and then exit the ML menu and start recording, you can press the ML button during recording, and it overlays the ML menu, with the selector still over the raw module, showing the current record speed. It doesn't seem to be as accurate as the recording indicator that pops up when recording though.

For reference with an estimate of around 163mbps from the ML menu, I can record continuously at 12bit. In case anyone else records 12 bit regularly. Actual speed is around 135mbps. I haven't experimented with the quality 10bit produces, but I am going to test that next since 10bit seems to be continuous in almost all exposure scenarios.

garypayton

  • Just arrived
  • *
  • Posts: 1
Re: Danne's crop_rec_4k & ISO experiments
« Reply #662 on: September 16, 2020, 02:37:10 AM »
Maybe this is the silliest question here, but when i try to install this build it tells me that i cannot install it because my firmware is not 1.2.3 ... I thought the right version of the firmware was the 1.1.3 (the one i have installed)...

What's wrong?

IDA_ML

  • Hero Member
  • *****
  • Posts: 885
Re: Danne's crop_rec_4k & ISO experiments
« Reply #663 on: September 16, 2020, 06:16:12 AM »
I haven't experimented with the quality 10bit produces, but I am going to test that next since 10bit seems to be continuous in almost all exposure scenarios.

In my experience, the only downside of 10 bits becomes obvious when you try to push severely underexposed shadows.  This results in ugly red-brownish noise in those areas, more obvious in the EOS-M.  Exposing to the right greatly reduces this phenomenon which is less pronounced at 12 bits.  Other than that, with normally lit scenes, 10 bits provide great results.

But please test yourself and share your findings here.

IDA_ML

  • Hero Member
  • *****
  • Posts: 885
Re: Danne's crop_rec_4k & ISO experiments
« Reply #664 on: September 16, 2020, 06:19:47 AM »
resolve does 18fps timelines so that would be a great choice!  i use it sometimes instead of 23.976 and it is really nice.

Glad to hear you like 18 fps too.  I love the smooth motion blur it provides.

IDA_ML

  • Hero Member
  • *****
  • Posts: 885
Re: Danne's crop_rec_4k & ISO experiments
« Reply #665 on: September 16, 2020, 12:17:36 PM »
I think, I have another good news:  The September 14-th build works quite nicely with cheap slow SD-cards without overclocking but with Card spanning active.  I tested with a 32 GB no name card that provides 39 MB/s write speed, tested with a card reader.  And I found that the 5D3 works with that card just as well as with the 95 MB/s Sandisk Extreme Pro.  I am getting:

1) 3264x1824/10bitLL/24fps, 0.0 exposure - continuous (orange indicator)

2) 3264x1824/12bitLL/24fps - about 10 sec. recording

3) 1920x3240/12bitLL/15fps/full res. anamorphic - continuous (green indicator)

Since the write speed on the SD-card indicator was well below 39 MB/s with both cards in the above modes, I think, we could safely use cheap slow cards with no or negligible write speed reduction.  In addition, we have another bonus - 10 to 20% more recording due to the spanned writing on both cards.  Not bad at all!

Quentin

  • Member
  • ***
  • Posts: 148
Re: Danne's crop_rec_4k & ISO experiments
« Reply #666 on: September 23, 2020, 05:21:16 PM »
I use 5d mk3 1.1.3.
I installed the last build 14 Sept 2020.

Somehow I get overexposed yellowish recordings.
I use MLVApp to develop the MLVs.
I drop the Gamma to 2 and select the best profile that satisfies me.
Before that build, I never had such distorted recordings.

Danne

  • Contributor
  • Hero Member
  • *****
  • Posts: 6965
Re: Danne's crop_rec_4k & ISO experiments
« Reply #667 on: September 23, 2020, 05:30:51 PM »
How do you think we could help you out? Impossible mission. At least throw in a sample mlv.

Quentin

  • Member
  • ***
  • Posts: 148
Re: Danne's crop_rec_4k & ISO experiments
« Reply #668 on: September 23, 2020, 05:41:16 PM »
How do you think we could help you out? Impossible mission. At least throw in a sample mlv.

Οκ, I will

Quentin

  • Member
  • ***
  • Posts: 148
Re: Danne's crop_rec_4k & ISO experiments
« Reply #669 on: September 24, 2020, 06:00:21 AM »
How do you think we could help you out? Impossible mission. At least throw in a sample mlv.

https://drive.google.com/file/d/1Z2hTLI5ETIbpyMDaOL3iWy8F9Eeeph9_/view?usp=sharing

sm105

  • New to the forum
  • *
  • Posts: 28
Re: Danne's crop_rec_4k & ISO experiments
« Reply #670 on: September 24, 2020, 07:26:05 AM »
https://drive.google.com/file/d/1Z2hTLI5ETIbpyMDaOL3iWy8F9Eeeph9_/view?usp=sharing

Seems like an issue on your end. It looks pretty good to me. The fluorescent lights are clipped, but everything else looks great. If it's yellow, maybe you have Night Shift on your display? ;)

MLV App:
Exposure -1.5
Temp/Tint 3740/41 (selected white area of wall that wasn't blown out)



Danne

  • Contributor
  • Hero Member
  • *****
  • Posts: 6965
Re: Danne's crop_rec_4k & ISO experiments
« Reply #671 on: September 24, 2020, 07:44:49 AM »
Looks perfectly fine to me.

Quentin

  • Member
  • ***
  • Posts: 148
Re: Danne's crop_rec_4k & ISO experiments
« Reply #672 on: September 24, 2020, 07:50:43 AM »
Seems like an issue on your end. It looks pretty good to me. The fluorescent lights are clipped, but everything else looks great. If it's yellow, maybe you have Night Shift on your display? ;)

MLV App:
Exposure -1.5
Temp/Tint 3740/41 (selected white area of wall that wasn't blown out)



What I was seeing on the monitor was normal, otherwise I wouldnt record it.
What was recorded was far away.
Sorry, but I will return to a previous build to prevent any risks.

Danne

  • Contributor
  • Hero Member
  • *****
  • Posts: 6965
Re: Danne's crop_rec_4k & ISO experiments
« Reply #673 on: September 24, 2020, 08:03:07 AM »
If there's nothing wrong it's always better to use the latest build. Usually bugs and other refinements have been worked on but do as you want.

wib

  • Freshman
  • **
  • Posts: 71
Re: Danne's crop_rec_4k & ISO experiments
« Reply #674 on: September 24, 2020, 10:40:46 AM »
exported as DNG from MLV app without any settings, and In Davinci a lut is applied and of course in the raw setting you can minimize the highlights and then you're done. the shot is perfect.


Here are the settings I modified in davinci.

project setting / camera raw tab :
raw profile cinema DNG
decode using project
color space blackmagic design
gamma blackmagicdesign
highlights recovery checked

I applied a lut. I used one that I created but you can use EOSHD LUT 5D3 RAW

in the color tab, in the camera raw menu :
color temp 3740
expo -1,5
sharpness 0
highlights -100
EOS 5D3 113 crop_rec_4k_mlv_snd_isogain_1x3_presets_2020Sep14.5D3113