Author Topic: Danne's crop_rec_4k experiments for EOS M  (Read 42687 times)

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2675 on: September 26, 2020, 11:40:03 AM »
I'm thinking if the 5K Anamorphic should be called 5K 1x3 instead or something similar. Newbies might think you need Anamorphic glass for that mode. But really pleased to see the LV Fix, looks excellent 👌
It´s internal anamorphic on sensor level. Rescaling/upscaling done in post anamorphically but I get why there could be issues understanding. Newbies won´t understand concept of 1x3 either though.

2blackbar

  • Senior
  • ****
  • Posts: 477
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2676 on: September 26, 2020, 10:15:32 PM »
I was just thinking, with the new 1920x1080 mode that reads evey pixel and is cropped into sensor , we have live preview now so... wouldnt it be better to replace that old 3x crop mode (that is enabled when you press "set" and you get 1x1 pixel sensor readout  ) with it to get full HD 1080p? Cause i remember asking about it in the past if it possible to get higher resolutions of 1080p when pressing "set" in HD 1080 mode , now its here, and id prefere to have full 1920x1080 instead of 1736x976 or whatever it is when cropped into 1x1 pixel redout.
What do You think Danne ? Or new 1920x1080 isnt stable enough ?

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2677 on: September 27, 2020, 09:31:45 AM »
We'll see.

rev787

  • New to the forum
  • *
  • Posts: 6
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2678 on: September 27, 2020, 01:54:31 PM »
Using Sep.26 build, after power off if I click display button menu button or d-pad then orange led will blink.

bern047

  • New to the forum
  • *
  • Posts: 10
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2679 on: September 27, 2020, 02:11:51 PM »
Nope mine doesn't blink, everything seems fine to me

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2680 on: September 27, 2020, 09:18:34 PM »
Friends -

I've been away from this forum for almost a year, but yesterday, I bought another M1 for use with my C-mount lenses. So far, I've always used the 3x crop method (available in the 2018 build), but I hope if you excuse me for asking this:

Is there now a way of getting 1920x1080 video with 1:1 sensor pixel format, i.e. not scaled? (I understand the 3x crop is actually scaled up a bit?)

(Probably answered somewhere else, but buried in lots of other posts... )

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2681 on: September 27, 2020, 09:47:48 PM »
Is there now a way of getting 1920x1080 video with 1:1 sensor pixel format, i.e. not scaled? (I understand the 3x crop is actually scaled up a bit?)
mv1080p will always be 3x3 binning. x3 or x5 zoom will be 1:1 sensor pixel format(cropped). Lucky we have the perfect compromise which is 1x3 binning(annamorphic modes). Lately even luckier as theBilalFakhouri outlined the necessary regs for correct realtime preview in anamorphic mode:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg231216#msg231216

Now I recommend test through all presets and read backwards about a 100 pages. Preview and exporting I´d recommend Mlv App.

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2682 on: September 28, 2020, 09:05:13 AM »
Thanks for the quick reply - the addy you provided gave a 404 error, but I loaded the Sep. 26 version and tried it out.

I probably didn't make myself clear in my original question; I'm not looking for an anamorphic or cropped version, just a 16:9 video using 1920x1080 (i.e. not 1736x976) not-binned 1:1 pixels directly from the sensor, in the standard 8-bit mov format with up to 30 min. recording.  (Got an interesting crash using the test version; camera locked, had to remove battery... :( )

Am I correct in assuming that this is currently not possible? I see that 2blackbar asked about it 2 days ago, but the current test is a different format, right? I don't want to use any other apps than Final Cut 7...

(Sorry if I'm asking stupid questions, but being away for almost a year, I'm virtually a newbie, again...  :P)

Here's the crash log:

ASSERT: 0
at ./MovieRecorder/MovRecState.c:415, MovieRecorder:ff1f3cb0
lv:1 mode:3

MovieRecorder stack: 1a7758 [1a7878-1a5878]
0xUNKNOWN  @ c380:1a7870
0xUNKNOWN  @ 22194:1a7848
0x00021E8C @ ff0e46c0:1a7830
0xUNKNOWN  @ 21ebc:1a7820
0xUNKNOWN  @ 21f44:1a7800
0x00001900 @ ff1f3cac:1a7790
0x0009E558 @ 9e864:1a7758

Magic Lantern version : Nightly.2020Sep26.EOSM202
Mercurial changeset   : 543f8692f847 (crop_rec_4k_mlv_snd_raw_only) tip
Built on 2020-09-26 07:24:36 UTC by daniel@MacBook-Pro-som-tillhor-Daniel.local.
Free Memory  : 221K + 2588K

("MacBook-Pro-som-tillhor-Daniel" - talar du också svenska? :) )

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2683 on: September 28, 2020, 09:27:46 AM »
@blackbar is asking about a refined version of x3 crop which allows for 1920x1276 cropped image on eosm(1920x1280 on 700D). Still cropped 1:1.
I still don´t know exactly what you are talking about but either you utilize the full sensor like when snapping CR2 files getting 1:1 but of course you will be able to do around 1fps movie wise or live with compromises, anamorphic, almost recording full sensor anamorphically.
If still not clear I recommend a month full time practical testing on your own, drawing own conclusions.
Error messages will mostly be impossible to follow if not including exact steps and settings applied before the error.

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2684 on: September 28, 2020, 09:43:35 AM »
Thanks again for quick reply. Sorry for not making myself clear - I'm almost 70 years old, and not as bright with software as you gurus here...

Basically, what I'm looking for is a simple 3x movie crop, 25 fps, standard mov format, just like the one available in the 2018Jul03 version, but a 3x crop with 1:1 pixels that is not scaled from a slightly smaller area (1736x976 ?) of the sensor - as I understand, that's what happens in the 2018 version?

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2685 on: September 28, 2020, 09:50:28 AM »
I see.

If you are after movie crop mode in h264 mov you can do following.

1 - Tap your screen and select h264 8bit preset:


2 - While in this mode push SET button and it will enter x3 crop. Enter it again to get back into regular mv1080p.

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2686 on: September 28, 2020, 05:05:56 PM »
I set and tested "h264 8bit", but the crop [/img]video is still the same as in the 2018 version, which, AFAIK, does not use 1:1 pixels from the sensor, but uses a somewhat smaller area of the sensor (1736x976 or whatever, as 2blackbar said on Sep26), and then scales it up to 1920x1080, losing a bit of definition in the scaling.

If this new version would give true 1:1 pixels, then it would give a slightly wider view - which it does not, it's exactly the same as in the 2018 version. I shot both a still frame and a 3x crop movie from the exactly same position, and the movie is about 10% enlarged when viewed full size compared to the still. This corresponds to the "1736x976 or whatever" crop from the sensor.

What I'm hoping for is a true 1920x1080 movie crop with 1:1 pixels from the sensor, in standard h264 8bit format, without any scaling. That feature would be most useful for anyone wanting to use C-mount lenses, and wanting the best possible quality, and still shooting up to 30 min. consecutive video.

Am I totally off track here?

(I don't need all the other esoteric multi-K anamorphic formats that only can record a few seconds max...  :o )


 

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2687 on: September 28, 2020, 05:51:45 PM »
Sounds pretty much off track yes.

Walter Schulz

  • Contributor
  • Hero Member
  • *****
  • Posts: 7941
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2688 on: September 28, 2020, 06:08:59 PM »
Danne: Is EOS M doing "native" H.264 Full-HD the same way as all other ML supported cams (5D3 excluded)? Horizontal binning and line skipping (only every third line used)? Or is the odd one here again?

Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2689 on: September 28, 2020, 06:11:20 PM »
H264 mov the same as 100d, 700d etc.

masc

  • Contributor
  • Hero Member
  • *****
  • Posts: 1816
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2690 on: September 28, 2020, 06:28:01 PM »
..., in standard h264 8bit format, without any scaling. ... wanting the best possible quality, and still shooting up to 30 min. consecutive video.
(I don't need all the other esoteric multi-K anamorphic formats that only can record a few seconds max...  :o )
The RAW modes are mostly continuous and offer the best quality you can get (way better then H.264 ever will be). 1920x1080 records without problem continuously. All you need is about 80GB at 1080p for 30min.
5D3.113 | EOSM.202

Walter Schulz

  • Contributor
  • Hero Member
  • *****
  • Posts: 7941
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2691 on: September 28, 2020, 06:41:06 PM »
Thanks, Danne!

@Janke:
As Danne wrote you are very, very lost!

Let's see what is in for you ...

1) All ML supported Canon cameras (apart from 5D3) do not record Full-HD (as in "Canon's H.264 implementation") by using 1920x1080 pixels. They use the full sensor width and height. But *not* every single pixel is used.
Horizontal pixels are binned = combined/processed resulting in something we call "native" resolution. This is (give or take)
Horizontal resolution for stills / 3 = native x resolution.
And that's always < 1920 pixels.
And lines are skipped. Only every 3rd line is used for recording H.264. And therefore "native" y resolution is about
Vertical resolution / 3 = native y resolution.
Do the math: It won't give you 1080 pixels.
To achieve Full-HD Canon does some scaling up. Software based softening applies to fight aliasing not being able to be surpressed by hardware AA-filter designed for still photo pixel pitch.

2) Canon introduced a crop mode for 600D. You press loupe/magnify in video mode and display will be filled but only a cropped area of the sensor is used. Without line skipping. Now guess how small the area might be ...

And ML found the code for that crop mode in later models 100D, 650D, 700D, EOS M and enabled it ... see feature matrix -> Movie -> Crop_mode_hack.

And that is pretty much all you can get for the foreseeable future. Take it or leave it or go raw/MLV.

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2692 on: September 28, 2020, 07:38:25 PM »
Thanks, Walter - now I'm beginning to understand (I hope...  :P )!

For Canon's own full-frame HD video, there's first horizontal binning & line skipping (which explains the aliasing), resulting in 1728 x 972, then upscaling that to 1920x1080, saved as h264.

In the following, I'm only talking about the 3x video crop:

It is already in the Canon firmware (but not implemented by Canon, only used by ML) so it apparently uses the central 1728 x 972 sensor pixels (no line skipping or horizontal binning as in "normal" HD). Then, that 1728x972 is again scaled to 1920x1080, and this is what is saved as h264. 

So the "simple" 3x crop is possible thanks to ML just utilizing the existing Canon firmware routines?

Now I understand that to get a true 1:1, 1920x1080 crop from the sensor, ML has to do its own procedures, and then save as raw.

Since I often shoot 2-hour jazz concerts (usually with three EOS-M cameras, one of them with a C-mount zoom), there would be massive raw files to convert before editing, so I was hoping for something more direct...

But, you can't get all you wish for (unless you're Jiminy Cricket...  :P )

Here is a sample from the latest, "pre-covid" concert, shot with one M1 in 3x crop mode and C-mount lens for the closeups, and for the wider views, one M3 and one M50 with Canon lenses (available in Full-HD):






Danne

  • Developer
  • Hero Member
  • *****
  • Posts: 7139
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2693 on: September 28, 2020, 08:03:29 PM »
Nice music.
x3crop is 1:1 pixel and as good as it gets if you want to use H264. Good lighting and correct white balance would work nicely for this but only 8bit. If you do raw there are all kinds of modes implemented by a1ex and further refined and tested working pretty good nowadays.

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2694 on: September 28, 2020, 08:12:46 PM »
x3crop is 1:1 pixel

Are you 100% sure? I have compared, in actual size, a still cropped to 1920x1080 and a 3x crop video shot from the same position, and the 3x video is about 10% larger = so it looks like it's scaled up from 1728 x 972, i.e. not 1:1 pixels.

(And yes, the venue lighting is for the audience, not for the videographer... :( )

Walter Schulz

  • Contributor
  • Hero Member
  • *****
  • Posts: 7941
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2695 on: September 28, 2020, 08:21:55 PM »
1:1 = Every pixel in a given area is used. Horizontal and vertical.
And the area is 1728x972 which is native resolution.
Therefore crop factor is 3x.

1:1 does not mean "Full sensor area" and does not mean 1920x1080!

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2696 on: September 28, 2020, 08:46:31 PM »
1:1 = Every pixel in a given area is used. Horizontal and vertical.
And the area is 1728x972 which is native resolution.
Therefore crop factor is 3x.

I've always thought 1:1 means "one sensor pixel is one pixel on screen" when viewed in "actual size", but that's not true for 3x crop video, since it is up-scaled from 1728x972 to 1920x1080 in-camera.

Walter Schulz

  • Contributor
  • Hero Member
  • *****
  • Posts: 7941
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2697 on: September 28, 2020, 08:53:17 PM »
Yes and no, 1:1 means "no pixel left behind in a given area" and therefore one pixel on sensor = 1 pixel on display. Your display resolution is not 1920x1080, it is "native resolution" 1728x1152 for stills and 1728x972 for 16:9 ratio.
LiveView resolution = Native resolution (with regard of x-y ratio)

Janke

  • Freshman
  • **
  • Posts: 80
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2698 on: September 28, 2020, 09:12:19 PM »
I think 1:1 means 1 sensor pixel = 1 display pixel. 

We're obviously talking of different things, I'm talking about watching a 3x crop video on a display of 1920x1080 size, such as a HDTV.

In that case, one sensor pixel is larger than one display pixel, i.e. there is not a 1:1 correlation.

With a "true" 1920x1080 pixel sensor video camera (such as Canon C300) one sensor pixel is one display pixel on a 1920x1080 size display.

Also, when you watch a still photo in "actual size" on a computer display (at least it's called that on a Mac), there is a 1:1 correlation between camera sensor pixels and display pixels.

Walter Schulz

  • Contributor
  • Hero Member
  • *****
  • Posts: 7941
Re: Danne's crop_rec_4k experiments for EOS M
« Reply #2699 on: September 28, 2020, 09:18:44 PM »
1:1 in ML terms means there is every pixel in a given sensor area used. Period. It doesn't say anything about the tv/monitor display you are using and which resolution it has. It simply doesn't mean anything to us. Stop trying to make that connection. There is none. Not for recording.
Canon and ML store data from sensor to card. It doesn't matter one bit for this processing chain if you have a monitor or not. We are talking about which parts of the sensor are used and how they are used (or not used).