Full-resolution silent pictures (silent.mo)

Started by a1ex, July 01, 2014, 05:11:15 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

a1ex

You can just merge 5D3-123 with fullres-silent-pics.

If you want to merge 3 branches, you can create a "work" branch, let's say starting from "unified", then merge 5D3-123 into it, then commit, then merge fullres-silent-pics into it.

More info: www.magiclantern.fm/forum/index.php?topic=9524

marekk

Is it normal that DNG fullres silent pic is much brighter than actual picture displayed in LV ?

a1ex

Answered in the first post and many times throughout the thread.

barepixels


marekk , exposure overide need to be on.  as instucted by alex
5D2 + nightly ML

ayshih

Quote from: a1ex on July 27, 2014, 09:03:01 AM
LiveView clocks seem to be a little slower:

5D3: 440 / 24.0 / 2080 =  8.81 ns/px, overclocked 7.97
60D: 546 / 24.0 / 1880 = 12.10 ns/px, overclocked 11.88
50D: 696 / 28.8 / 1664 = 14.52 ns/px, overclocked 14.31

I think you made a typo: doesn't the 60D have a 28.8 MHz clock (as you used earlier)?  I think it should be:

60D: 546 / 28.8 / 1880 = 10.08 ns/px, overclocked 9.90

which is closer to the photo readout time.

I wonder how much time is needed to bin pixels, and perhaps the 50D's hardware can't do it efficiently.  Comparing the 50D and 60D in LV zoom mode:

50D: 636 / 28.8 / 2064 = 10.70 ns/px, overclocked 10.60
60D: 734 / 28.8 / 2520 = 10.11 ns/px, overclocked 10.09

which is a lot closer than in LV normal mode.
Canon EOS 50D | 17–40mm f/4L & 70–300mm f/4.5–5.6 DO IS | Lexar 1066x

a1ex

Correct.

From the rolling shutter tests in video mode, 5x zoom and 1080p use roughly the same timing on 60D.


60D 1080p  : 546 / 28.8 / 1880 = 10.08 ns/px, overclocked 9.90
60D 5x zoom: 734 / 28.8 / 2520 = 10.11 ns/px, overclocked 10.08

sdesign

my 5d3 still has brightness issue even with exposure override on

a1ex

Quote from: a1ex on July 27, 2014, 03:33:10 PM
Answered in the first post and many times throughout the thread.

and no, it's not about exposure override.

wyrlyn

if I'm right the overexposing is caused by the 1/10sec limit, but you can get rid of that by using fader ND or just post-process the file using Camera Raw (adjusting exp).

a1ex

Right, and the exact timings for the fastest possible exposure were analyzed in the previous pages (all those funky graphs).

a1ex

Quote from: a1ex on July 25, 2014, 08:12:53 PM
I'll merge the full-res silent picture branch next week, and I'll break all the nightly builds for which I don't have a raw_diag screenshot.

Hint for 6D, 7D, 550D, 650D, 1100D and EOS-M.

Walter Schulz

Non-coder with 650D and 7D here. Anything I can do?

a1ex

Not really. Sadly, breaking the builds seem to be the only way to "motivate" some people, so I'm going to use it more often (better breaking the builds now, instead of discovering it's totally broken after 5 months or more - see the 600D screenshot, for example).

Levas

Quote from: a1ex on July 25, 2014, 08:12:53 PM
I'll break all the nightly builds for which I don't have a raw_diag screenshot.

Noooooooo... Quick, where's my 6d  ;D

I downloaded the raw_diag.mo and made the screenshot(see PNG) you asked for(plus some offset info):
https://drive.google.com/folderview?id=0B1BxGc3dfMDaazVGRF9JOTF2VzQ&usp=sharing

It's not 100% perfect aligned as you can see, made this with the 18Juli2014 Nightly build for 6d.
The offset used in this Magic Lantern build seems, Active area: 50, 84, 3708, 5554 (crop size 5470,3658).

The offset seen in exif info from normal Canon CR2 is:
Sensor Left Border = 84
Sensor Top Border = 50
Sensor Right Border = 5555
Sensor Bottom Border = 3697
Which results in an image 5472 width and 3648 Height


Hope this is enough data to keep the 6d alive in the nightly builds  :D

a1ex

Quote from: Levas on July 30, 2014, 08:13:49 PM
Hope this is enough data to keep the 6d alive in the nightly builds  :D

Nope, it should be corrected. The test should be done with the fullres-silent-pics branch (the offsets changed).

With these results, ETTR on this camera is also not much better than random exposure.

Levas

I'm no coder, I have no idea how to correct this ?

To do this test with the fullres-silent-pics branch, I probably need to compile I guess  :-[ (which is beyond my knowledge...)
Or is there somewhere some place where there is a build for the 6d with fullres-silent-pic branch ?

Audionut and NikFreak, please, help is needed over here, I guess...

Levas

@Alex,

Tried the raw_diag.mo (downloaded it today) with the newest Nightly build for Canon 6d.
But it doesn't work, it says in the module tab:
Wrong version, (v5.0, Expected v6.0)

dmilligan

1100D uses edmac channel #2



first picture works fine, take another and I get Err80 (no logs).

EDIT: the error seems to be related to dumping the raw buffer as DNG (I guess this is b/c dng code modifies the raw buffer and it looks like it's a little too big)

a1ex

That's quite possible (had this problem on 5D3). You could reduce height a little - check EDMAC channel 2 in Debug menu to find the exact size. You may have to add 1 to the height displayed there (upload the EDMAC screenshot and I'll tell you).

Nice fonts :D

For 6D, a good starting point for the offsets is here: https://bitbucket.org/hudson/magic-lantern/commits/e261cbbbdf25a380f74c15ddb671952518693b26#comment-1089535 (reduce the resolution to be able to read everything).

dmilligan

reduced the buffer start offset by 3 rows and fixed the skip offsets, no more crash when saving DNG


        #ifdef CONFIG_1100D
        //  from debug log: [TTJ][150,2551,0] RAW(4352,2874,0,14)
        width = 4352; //From TTJ Log
        height = 2874;
        skip_top = 16;
        skip_left = 62;
        raw_info.buffer += width * 14/8;
        #endif



dmilligan

Quote from: a1ex on July 30, 2014, 11:11:45 PM
Nice fonts :D
They look even better nearest neighbor on the tiny 1100D screen :P

a1ex

Do you still need to skip one line? (check raw zebra colors)

If it's like this, it's correct:


If it's like this, it's wrong:

dmilligan

I checked the DNG colors, they are correct.

homeros

I'm not expert but I was wondering. Is it transfer speed to card that preventing record in higher FPS or just sensor won't let it? If it is transfer speed then can be pictures shrink to low resolutions with bicubic or bilinear interpolation (Instead of line-skipping) in camera then save to card. With this we can eliminate moire and aliasing at least mostly. It might be discussed earlier or not. Because it might be stupid idea, I don't know.

There is one more idea that might be stupid. Can HDMI connection be used for transferring raw data like Alexa's T-Link? As you know Alexa uses dual link SDI as a data connection to Codex. May be HDMI connection can use for something like that. HDMI 1.0 has 5.0 Gbit bandwidth in theory. 5120x2700 (5K) 14 bit 24p raw fits in that limit. I know still need a compatible device that hasn't been invented yet. Just ideas...

nikfreak

@Levas:

For 6D I compiled for you "fullres-silent-pic" branch but before compiling I made changes to "raw.c" as a1ex and 1% have been discussing on bitbucket:

First I defined edmac like discussed above:


#if defined(CONFIG_5D3) || defined(CONFIG_700D) || defined(CONFIG_6D)
#define RAW_PHOTO_EDMAC 0xc0f04008
#endif


and also changed height + skip-values according to a1ex's suggestion on bitbucket discussion linked above:
#ifdef CONFIG_6D  //Needs check from Raw dump but looks aligned.
width = 5568;
height = 3722;
skip_left = 72; //Meta Data
skip_right = 0;
skip_top = 52; // Meta Data
#endif


Here's the zip to try for you @Levas:

http://www59.zippyshare.com/v/38767089/file.html

Please lemme know if you have no time to do the necessary test before a1ex will drop 6D out of this


Before anyone gets the idea to use thos for anything othe rthan 6D: I had to use fake stubs and defines for some cameras to get this compiled. So this will only work with 6D.


Edit: fixed link with correct zip containing modules
[size=8pt]70D.112 & 100D.101[/size]