Author Topic: Danne's crop_rec_4k experiments for EOS M  (Read 239337 times)

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #75 on: January 07, 2019, 01:41:51 PM »
I see. Disabled audio and global draw too?
Thanks for testing.

loknar

  • Freshman
  • **
  • Posts: 78
  • 550D, EOS M
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #76 on: January 07, 2019, 01:56:32 PM »
I did some more tests
i'm getting ~250 frames without SD overclock (~35MB/s), ~ 550 with SD overclock (~44MB/s)
in second set corruption appeared only in shots with SD overclock (3 ouf of 550). Frame out of order didn't appear again.
As for global draw it's on, but i'm using canon liveview, does it count?
Disabled sound, like not loaded mlv_sound module?

loknar

  • Freshman
  • **
  • Posts: 78
  • 550D, EOS M
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #77 on: January 07, 2019, 02:03:17 PM »
Aaaand some more tests.
When not overclocked no corrupted frames.
Sound doesn't seem to cause problems.

Also i noticed it doesn't want to show dowm when button pressed (it stucks on "Sensor cleaning" and orange led on), needs two additional presses (one to start again asnd switch off)

With SD overclock and no sound it seems to prolong recording to some ~800 frames and no corruption
Sorry about first report, i  enabled sound module on autopilot.

Also, all these are low light situation no Sun here today.

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #78 on: January 07, 2019, 02:48:34 PM »
Thanks for tests. New upload:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

Ran five 11 sec recordings with the 50 fps preset and couldn´t create a corrupted frame. Maybe fixed?

loknar

  • Freshman
  • **
  • Posts: 78
  • 550D, EOS M
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #79 on: January 07, 2019, 03:33:13 PM »
It now switches off okay.

For some reason i'm getting longer times 650-800 frames but unfortunately each clip contains 3-5 corrupted frames, not consequent in different parts of clips.

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #80 on: January 07, 2019, 05:29:33 PM »
Ok, final attempt for corruption free files in 50fps:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

loknar

  • Freshman
  • **
  • Posts: 78
  • 550D, EOS M
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #81 on: January 07, 2019, 05:54:15 PM »
No corruption and length ~17 seconds shots, I only activated SD overclock, cropmode and set 10 bits in it.

Doesn't like overexposure and sudden changes in brightness, when light conditions don't change too much, it looks like it can be continuous (I have 21 second shot that I had to direct at light bulb in the end to make it stop [after several seconds]).

Btw does ML control over Image Stabilization in lens? I have IS activated in Canon menu, but in replay it says "no IS" (i use manual focus, but that shouldn't be a problem or is it?)

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #82 on: January 07, 2019, 07:08:45 PM »
Cool.
No idea about image stabilisation in this case.

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #83 on: January 09, 2019, 11:46:32 AM »
New build:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

Optimized 2520x1308 setting to 2520x1438(16:9). Works with real-time preview 24fps.

Also added a mv720p 50fps setting. Will record 1736x696 so unstretches in post to 16:9.

Pacerx999

  • New to the forum
  • *
  • Posts: 12
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #84 on: January 09, 2019, 02:18:33 PM »
So utilising crop_rec_4k_mlv_snd_isogain_1x3_presets_2019Jan07.EOSM202

I manage to get both 50fps with mv1080 and the 4k-24fps.

With the 4k-24fps, I was getting a similar colour artefacts to Loknar. I.e. blue/yellow oddness. Seems to be concentrated around horizontal lines

Also overexposure or anywhere close gets pink highlights when exposure is brought down. Highlight reconstruction in MLVApp doesn't help, but Resolve highlight reconstruction does help, but only using MLVFS, which currently doesn't remove focus pixels.

Obviously preview is slow, but finding and maintaining focus is v. difficult.

With all that said, when you get the exposure right (ETTR) and grab focus the quality of the shot is excellent, but the horizontal artefacts are affecting things. Just want to be clear, I think it's Amazing to get 4k RAW out of a 6year old camera, but would like to get this to a great user experience so it could be used for actual project.s

50Fps seems to be working fine, with SD Overclock, I'm getting approx 10sec, but it was a bright scene, so I think that's what limits it due to data rate. But it looks good when slowed down to maybe 80/90fps in Premiere using optical flow.

I see that there is a Jan9 build, so will go on to try that and report back any findings. If MLVs would be helpful for the colour artefacting, just let me know

TA!

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #85 on: January 09, 2019, 02:22:00 PM »
4k is hard. Compromise is it runs 3x1 binning(mv720p) which is, well, not 3x3 binning. Useful or not I tend to keep prioritized presets first and then the longer down the list the crazier it can get  :P

masc

  • Contributor
  • Hero Member
  • *****
  • Posts: 2011
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #86 on: January 09, 2019, 04:28:17 PM »
Also overexposure or anywhere close gets pink highlights when exposure is brought down. Highlight reconstruction in MLVApp doesn't help, but Resolve highlight reconstruction does help, but only using MLVFS, which currently doesn't remove focus pixels.
You should be able to correct pink highlights in MLVApp too: correct first RAW White Level. -> activate Highlight Reconstruction and then lower RAW White Level until pink highlights go away.
5D3.113 | EOSM.202

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #87 on: January 09, 2019, 06:22:22 PM »
New build:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

Added a 5K 24 fps (squeezed 2.35:1) preset into the game...

optical-work

  • New to the forum
  • *
  • Posts: 19
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #88 on: January 09, 2019, 07:30:17 PM »
as far as i can see 4k 3x1 mode has better quality than mv1080 mode and raw therapee's defringe tool gets rid of the color artifacts. but there is too much moire compared to mv1080.

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #89 on: January 10, 2019, 12:11:12 PM »
New upload. Again  8).
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

Finally got around to fix the cool 1x3 binning mode which until now was restricted to height stop around 1120. The preset now goes all the way to 1712x2184. Files opens up just fine in Mlv App. More info from this thread(@theBilalFakhouri) about 1x3 binning:
https://www.magiclantern.fm/forum/index.php?topic=22818.msg206134#msg206134

Comparison files:
https://bitbucket.org/Dannephoto/magic-lantern/downloads/1_x3_M10-1218.MLV
https://bitbucket.org/Dannephoto/magic-lantern/downloads/3x3_M10-1219.MLV

What eventually worked was bypassing any height restriction for eosm in raw.c and then start expanding registers. Commit:
https://bitbucket.org/Dannephoto/magic-lantern/commits/0b5f6fe465afe77de97d57255151f8644b067395

Also added some rearrangements and took the liberty to rewrite a few info parts.


Next up, 100D...

jacquaviva

  • New to the forum
  • *
  • Posts: 17
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #90 on: January 10, 2019, 11:07:18 PM »
Hi,

I am a bit confused about crop factors from all these new presets.
Are 5K, 4K & 1712x2184 still close to 3.33x crop ?

Is there still any hope about 1.6x crop or it always will be 1736x696 stretched to 1736x1160 ?

Thanks in advance.

By the way, since Canon M flange distance is 18mm the same than Sony E, I found how to build and adapt a Sony E mount part to the EOS M in order to use the Fotodiox ND Throttle Canon EF to Sony E (among others).
It works very well. I only use full sensor 1.6x RAW video and to avoid moire and aliasing, I have bought a 6D VAF Mosaic Engineering AA filter and achived to tinker with to adapt it to the camera. Working very good as well.
If anybody wants more information.

12georgiadis

  • Member
  • ***
  • Posts: 213
  • 5DmkIII - 7D - EOS-M
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #91 on: January 10, 2019, 11:23:25 PM »
to avoid moire and aliasing, I have bought a 6D VAF Mosaic Engineering AA filter and achived to tinker with to adapt it to the camera. Working very good as well.
If anybody wants more information.
Do you have some instructions and pictures to show us how to adapt VAF filters ? Sounds interesting =)

dfort

  • Guest
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #92 on: January 11, 2019, 02:19:07 AM »
Here's someone else who shoehorned a VAF filter into the EOSM. No instructions or photos. Hint -- Scotch tape.

Yes but it's on a cheap one. So you have more space to do it.
It's a bad work with scotch, I am not as proud of it to share pictures of it. Sorry.
You just have to put any kind of filter between the lens and the sensor and it's effactive...

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #93 on: January 11, 2019, 09:55:16 AM »
Is there still any hope about 1.6x crop or it always will be 1736x696 stretched to 1736x1160 ?
?
If you mean to record mv1080p 3x3 binning it´s already working.

loknar

  • Freshman
  • **
  • Posts: 78
  • 550D, EOS M
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #94 on: January 11, 2019, 12:16:11 PM »
Is it just me, or build crop_rec_4k_mlv_snd_isogain_1x3_presets_2019Jan10.EOSM202 is creating corrupted frames in 1x3 binning mode?

And this is probably stupid question, but instead of 3x3,1x3,3x1, would it be possible to use binning or line skipping 2x2...? Probably using some array operations, like readout whole sensor and then decrease it in size...not quite sure if limit is on processing side, or write speed size (or if it even possible to edit raw frame before saving it...but given how many "impossible" things have been already done...).

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #95 on: January 11, 2019, 12:23:35 PM »
Corruption is not unlikely. What were your exact settings? Will check for it.

Anyway, here´s an attempt in Mlv App using dualiso. Looks pretty nice:

Here´s a dualiso attempt coming from eosm and it´s 1x3 mode setting. Recorded in:
10bit,
sd_uhs,
24 fps,
dualiso 100/1600,
1600x2040(2.35:1)


mlv sample:
https://bitbucket.org/Dannephoto/magic-lantern/downloads/1x3_dualiso_eosm_M11-1209.MLV








Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #96 on: January 11, 2019, 12:52:35 PM »

jacquaviva

  • New to the forum
  • *
  • Posts: 17
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #97 on: January 11, 2019, 02:44:32 PM »
I mean what will be the max resolution for RAW video using full sensor read out (1.6x crop) ?
I know it's working with 1736x696, but not more for 24fps.

And what are the crop factors for new preset 4k & 5k ?

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7391
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #98 on: January 11, 2019, 02:56:53 PM »
Just try it yourself and draw conclusions. Don´t really get what you´re asking.

Walter Schulz

  • Contributor
  • Hero Member
  • *****
  • Posts: 8487
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #99 on: January 11, 2019, 02:57:44 PM »
@jacquaviva:
Crop factor calculation into 35 mm terms:

(Sensor horizontal pixel count / ML horizontal resolution in ML crop mode) x 1.61

It's not exactly rocket science, you know?