Danne's crop_rec_4k experiments for EOS M

[Danne]

Yeah, the line was easy to get rid of. We need to check what else is going on.
By the way. With this applied in adtg_gui I could correct the black border to the right. Having trouble applying it in crop_rec.c though:

Code Select Expand

case 0xc0f383d4: return 0x4f0010;
case 0xc0f383dc: return 0x42401c6;
Not always sure what regs are applied and which are not going to be accepted in there...

[gravitatemediagroup]

I did a quick 1 clip test of the new 1080 setting and my results were similar to the screen shot a few post up.  ISO was at 800, that's the only specific setting remember, I deleted already assuming it was something I had wrong, but I figured I'd report the issue letting you know it may be something in the build

[Danne]

I need mlv files and test comparisons since I can't reproduce the issue. You can reproduce.

[masc]

@Danne: this was filmed this morning when testing yesterdays build (Feb15). I really loved the quality of this version. It was really hard to find some moiree... (one roof had some). Hope we get this back soon without red line, without noise & moiree!

[Danne]

Here is a version which will help testing:
https://bitbucket.org/Dannephoto/magic-lantern/downloads/crop_rec_4k_mlv_snd_isogain_1x3_presets_2019Feb16.EOSM202TEST_rewire.zip

Once all is set and rewired you can enter Crop mode sub menu and set test_rewire to "old" for the settings with the thin red line at the bottom. If set to off you get the latest settings without the thin red line. We need static test scenes for comparisons of moiré and other anomalies.

[jacquaviva]

Thank you Danne for your work, correct preview is wonderful.
Nevertheless, I get a strange white cross pattern in the middle of each frame and more in general all over the frame a black pattern.



 

[dfort]

Those are focus pixels. They can be removed in post. Working on it.


Sent from my iPhone using Tapatalk

[masc]

Hi Danne, light is back, so I tried out your build now. But it looks not as expected:
-rewire=old: no red line
-rewire=off, -2+2=no red line
-rewire=off, +0-0/-1+1=red line
-rewire=off, +1-1/+2-2=red line + one white line
-rewire=off, +3-3/+4-4=red line + two white lines
-rewire=off, +5-5/+6-6=red line + three white lines
Moiree looks always identical (better than yesterday, but can't compare to Feb15). But I don't find stair effects.

[Danne]

Try refreshing after enabling "old" by opening canon menu and also maybe ml menu.
Tested this morning and it´s not possible to reproduce any issues. What happens is that aliasing can be produced differently but that has more to do with focus, not with ratio changes.

[masc]

Okay... got it reproduced, but not as expected. Will shorten the files and upload. Best by far is rewire=Off, -2+2.

[masc]

The only two settings without bottom line are rewire=old and rewire=off-2+2. But in rewire=old I get stairs, hard focus pixels and more artifacts. So in this build I prefere rewire=Off, reg_skip_top=-2, reg_skip_bottom=+2.

You'll find a png with some marked areas in the frame where I looked at details and my receipt file.
https://www.dropbox.com/s/wddfxe3p7zmdsn1/RewireTestrun.zip?dl=0

[jacquaviva]

Thank you Dfort. I can get rid of focus pixels by using the debayer amaze in MLVapp, but can't use it by exporting to dng sequence. How to ?

[Danne]

Set chroma smooth to 3x3.

[Danne]

@masc
Settings right now if test_rewire is set to off:

Code Select Expand

        skip_top = 84;
        skip_right = 60;
    skip_bottom = 0;

Let´s apply your optimised setting "So in this build I prefere rewire=Off, reg_skip_top=-2, reg_skip_bottom=+2.":

Code Select Expand

        skip_top = 82;
        skip_right = 60;
    skip_bottom = 2;

Only one problem. The first commit after noticing the thin red line was exactly the setting above:
https://bitbucket.org/Dannephoto/magic-lantern/commits/5cb143d56c2dcb9a16f27c5094e4f8f1ee8fb52b

Your comment to that commit:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg212152#msg212152

So, I revert back to the first commit as long as we all agree this problem is not as easy as we first thought? Any other thoughts?

Also note that when test_rewire is set to old it´s the exact lines as with the build before any thin redline fixes were applied(same as february 15th).

[masc]

Very strange. Do we have a kind of sign error somewhere? Don't understand all this at all.

[Danne]

Placebo?
Anyway. Here´s a build with your proposed settings working best:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

Here is the feb 15th build compiled exactly before the thin red line fix in commit tree(compiled today and renamed):
https://bitbucket.org/Dannephoto/magic-lantern/downloads/crop_rec_4k_mlv_snd_isogain_1x3_presets_2019Feb15.EOSM202.zip

[jacquaviva]

Thank you, chroma smoothing works with dng out.
Danne, is there a way to save crop mode activated when booting the camera. I always have to switch it on again. Thanks

[Danne]

Yes, you can enable it but I havn´t found a way to disable it when changing presets in crop rec once it´s enabled. In crop_rec.c put in this at the top:

Code Select Expand

#include "crop-mode-hack.h"

Then inside static int crop_rec_needs_lv_refresh(): in crop_rec.c put in following:

Code Select Expand

if (crop_preset == CROP_PRESET_mcm_mv1080_EOSM)
{
        movie_crop_hack_enable();
}

Compile.

[masc]

Placebo?
Anyway. Here´s a build with your proposed settings working best:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

Here is the feb 15th build compiled exactly before the thin red line fix in commit tree(compiled today and renamed):
https://bitbucket.org/Dannephoto/magic-lantern/downloads/crop_rec_4k_mlv_snd_isogain_1x3_presets_2019Feb15.EOSM202.zip

No, don't think so. The "stairs" are very easy to identify at 45° edges, same for the red line. But what happened for your "rewire=old" preset? Why activates it the stairs, and -2+2 doesn't. And all this does not fit to yesterdays build. Something must be wrong.

Will test again the posted builds.

[jacquaviva]

Thanks Danne, I don't know to compile but I'll figure it out.
Thanks you again for this work !

One question about MLVapp, is there a way to set by default 3x3 chroma smoothing, I have to set it for each MLV file.

[masc]

@Danne:
I expect indeed a sign error somewhere. I again tested the first build after the red line (the one with the stairs). Reproduced all I saw yesterday. BUT: reg_skip_top=-4 and reg_skip_bottom=+4 looks perfect
What happens for:

Code Select Expand


        skip_top = 86;
        skip_right = 60;
    skip_bottom = -2;

@jacquaviva
Activate it once for one clip, Ctrl+C (maybe activate in mask chroma smoothing only), Ctrl+A, Ctrl+V.

[Danne]

-4, +4, link or name of build used?
What's a "sign error"?

[masc]

Build used is: crop_rec_4k_mlv_snd_isogain_1x3_presets_2019Feb17.EOSM202
-4+4 looks perfect. Registers have 82 and 2 by default, you wrote

Build before:
-2+2 looked perfect. Registers were 84 and 0 by default.

So I think the math behind is not default+offset, it is default-offset. The algebraic sign (+ instead of -) should be wrong - or I am... confusing... 

So I would try 86 and -2 next.

[Danne]

Build used is: crop_rec_4k_mlv_snd_isogain_1x3_presets_2019Feb17.EOSM202
-4+4 looks perfect. Registers have 82 and 2 by default, you wrote

-4+4 equals:

Code Select Expand

        skip_top = 78;
        skip_right = 60;
    skip_bottom = 6;

86 and -2 as suggested would become

Code Select Expand

        skip_top = 86;
        skip_right = 60;
    skip_bottom = -2;
And then we have a fat white border at the bottom.

We need to shave at minimum 2 at the bottom to get rid of the bottom pixels:

Code Select Expand

    skip_bottom = 2;
Since default skip_top is 72 we can safely add back what´s shaved at the bottom since we here start at:

Code Select Expand

    skip_top = 84; so decreasing to 82 will keep us at 1736x976(16:9):

Code Select Expand

        skip_top = 82;
        skip_right = 60;
    skip_bottom = 2;

I cannot reproduce the issue described so I really suggest you download my branch and get right into crop_rec.c and compile crop_rec.mo with any optimised settings you come up with. It´s easy with Compiler.app. Just get into the modules area and find this menu:

[masc]

-4+4 equals:

Code Select Expand

        skip_top = 78;
        skip_right = 60;
    skip_bottom = 6;

...if the register screen does what we think it does...

86 and -2 as suggested would become

Code Select Expand

        skip_top = 86;
        skip_right = 60;
    skip_bottom = -2;
And then we have a fat white border at the bottom.

Have you tried out? Oh man... I really don't understand what happens here.

I cannot reproduce the issue described so I really suggest you download my branch and get right into crop_rec.c and compile crop_rec.mo with any optimised settings you come up with. It´s easy with Compiler.app.

What a shame it seems I am the only one getting this. Unfortunately I am here with OSX 10.9, which is not supported by your compiler app (or by brew).