Tragic Lantern for EOS M

Started by coutts, April 17, 2013, 01:43:28 AM

Previous topic - Next topic

0 Members and 2 Guests are viewing this topic.

jerrykil

Quote from: qsara on June 21, 2013, 11:18:55 PM
I'm using just vegas pro for HDR videos, works like a charm on 720p videos, at 1080p it is really hard to make it work with my GPU.

Very nice, qsara! Can I ask what CBR you are using? Is there a way to tell if i'm losing frames due to increased bitrate?
Thanks for sharing

haomuch

Hi it's very exciting to see the ml firmware for eos m. Just wondering if we can add screen off function in the next update for the 2.0 version firmware. The eos m currently doesn't provide evf but I would like to use a ovf for my 22mm and 50 mm lens. Therefore a screen off function would be very useful. Maybe we can make it by long pressing the info button.

feureau

BTW, the crash log from the creative filter crash:

ASSERT: IsSuiteSignature( hSuite )
at ./PackMemory/PackMem.c:577, task InnerDevelopMgr
lv:0 mode:2


Magic Lantern version : v2.3.NEXT.2013Jun15.EOSM106
Mercurial changeset   : 57cd362ac2f5 (unified) tip
Built on 2013-06-15 17:17:36 by user@D610.
Free Memory  : 247K + 2024K

1%

Ok, will have to check on this and see why... looks like a memory issue.

feureau


mixer2

oh, nice... i tested the latest version and now it's possible to write 36-37mb/s in videomode (without multishot nr), too.
any progress on the dot removal?

if there are linux users, we may discuss the used workflows. what i do is Raw2Dng via wine => convert dng to exr via darktable cli => unsqueeze exr with imagemagick mogrify => open exr in darktable for postprocessing => export to jpeg => import to video software.
i wrote a simple bash script for the first 3 steps to get unsqueezed exr frames from raw with a single command.

haomuch

Hi I am using the ML firmware for my EOS M with the FD 50mm f1.4 len. It's really a great tool for manual focusing. But I am wondering whether I can disable the addition information display on the shooting mode and just show the peak focusing? I think the stock display of Canon is already enough and the ML display will cover those Canon display and settings.

haomuch

Quote from: mixer2 on June 22, 2013, 10:09:55 PM
oh, nice... i tested the latest version and now it's possible to write 36-37mb/s in videomode (without multishot nr), too.
any progress on the dot removal?

if there are linux users, we may discuss the used workflows. what i do is Raw2Dng via wine => convert dng to exr via darktable cli => unsqueeze exr with imagemagick mogrify => open exr in darktable for postprocessing => export to jpeg => import to video software.
i wrote a simple bash script for the first 3 steps to get unsqueezed exr frames from raw with a single command.

Do you mean the latest ML alpha firmware for EOS M? I only find one in the OP~

1%

I think for dot removal we need to take both sizes and edit the 650D dot remover... unless someone knows how to add them to chdk-dng.c correctly.

Its on the list along with enabling actual shutters in eosM.. the ones right now aren't correct (not using shutter_timer.)

mixer2

i tested how to get the dot positions, it's not perfect, but this is the result so far:

it's a mask for 1600x542 (16:9)
i used this image as base, maybe anyone can find better settings to seperate noise from the dots.
http://dl.dropbox.com/s/mow4swn06j3ojck/000000-hdr.dng

to generate the image i recorded ~180 completely underexposed frames with 100 iso, then combined all to a hdr image to eliminate the noise. maybe i should try more frames to get even better seperation of frames to the noise. i've to check what you exactly have in the dng and how to mask the images with the result. do you have and documentation about what's exactly in the dng container?

1%

Try silent pic of the whole thing in normal and zoom (shouldn't have them, but check). They shouldn't change based on what you crop from the image.

mixer2

oh, that's correct, if we've the data for the maximum resolutions, we can just crop on the center for the resolution we need.

EDIT: played a bit around with the dotremover sourcecode (the logging lib dependency was missing -.- ) and tested the data from the png i posted above. it's not perfect so far but removes most of the dots. i've to check how to improve the removal data. i wrote a pm to foorgol and asked him to put the source into a repository. maybe he can also give me some tipps on how to improve the removal data.

Before:


After:



EDIT: updated the example images. in this images the dots are much better visible. marking the dots as dead pixels doesn't work for me. when i use this setting i get black pixels (what does make sense, if it's set to 0). are you sure, that setting it to 0 is the correct way? i don't know anything about dng, but the "FixBadPixelsList" in the specification looks more complex than just setting a pixel to 0.

EDIT: optimized the data which pixels have to be removed and updated the example images. the data is now for 1734x695 silent dng. this should work (with correct pixel offset) for all non zoomed video resolutions. the dot remover app has to be modified, that it handles the offsets correct for all resolutions. and there should be a setting to seperate eos m from 650d. i don't want to modify to much, till there is a repository. it would just be hard to merge the source together, without a version control system.
i think the data should work better than the result is, but the way the interpolation works isn't perfect so far. have to discuss that with foorgol, as soon as he responds to my pm.

EDIT: used the 1734x695 removal data for an 1600x542 image and it worked just fine. the correct offset for x is 70 and for y 76. i don't calculate them automatically so far. have to check how the center crop has to be caculated, becuse it's not completly in the center, but shifted few pixels because of the bayer pattern.

Before:

After:


EDIT: i asked in the dt irc how that dng setting to 0 stuff works and why those pixels are black in dt. if i understand it correct the FixBadPixelsConstant is set to 0 in the dngs generated by ml. that's why setting to 0 marks them as bad pixels.
dt doesn't seem to support dng 1.3.0.0, that's why the pixels are black.

EDIT: 1% can you explain me, why center crop of 1600 doesn't start at pixel (1734-1600)/2 = 67, but at 70? 68 would make sense for me , because maybe it should be even because of the bayer pattern. but i don't know why it should be 70. in the ml source there is some other stuff in the calulation like raw_info.active_area.x1. don't know what that exactly does. can you tell me how to reliably calculate the offset. is this possible with just the image size or do we need some additional information?

EDIT: did a small videotest.
http://dl.dropboxusercontent.com/s/lincex5op3relr7/dotremoval.mp4
- first is h.264
- secound is raw
- third is based on the same data as the secound but with dot removal

EDIT: one more question. is there any information in the header from wich camera the dng is? if not, may it be possible to add it?

1%

Header should say EOSM.. it does for 6D/5d3. Also don't forget frame skips... probably why coordinates are off.

[EOSM] Raw TypeTest.zip
https://mega.co.nz/#!o94RTLQZ!FHLUZkjExsOTg7_Rc8foRCJHEVMd1nHZ3Olt9j0I8Fs

Have another chance at raw types I was off by 0x4. See if any of them don't damage edges + have no dots/other defects.

mixer2

how do skipped frames change the crop offset of non skipped frames?
can you write me why exactly from the maximum resolution 1734x695 the horizontal offset is 70. 70px border 1600px image 64px border. how is it exactly calulated?

new raw type doesn't work well... i get edge detection and focus dots on this one. but... if there is one with edge detection, one with focus dots and one with both, maybe there is also one with none ;)

1%

You should get 100 of them to check. Test is on don't click me.

I think its total size - skips. So if you do raw video its a crop of that and relative to center?

haomuch

Quote from: feureau on June 22, 2013, 04:45:04 PM
BTW, the crash log from the creative filter crash:

ASSERT: IsSuiteSignature( hSuite )
at ./PackMemory/PackMem.c:577, task InnerDevelopMgr
lv:0 mode:2


Magic Lantern version : v2.3.NEXT.2013Jun15.EOSM106
Mercurial changeset   : 57cd362ac2f5 (unified) tip
Built on 2013-06-15 17:17:36 by user@D610.
Free Memory  : 247K + 2024K


OMG! Where can I download the June 15 version? Please tell me the address.

mixer2

ah, ok, i checked all the 100 images. 81 were completly distorted.
from the 19 remaining:
4 had af dots all over the image. we've to check if this will lead to better or worse quality after removal. the difference is, that the af pixels are just red and blue pixels instead of the other type, were most of them were on the green pixels. on the one hand bayer has already a lot more green than red and blue pixels, but on the other hand the green are more significant for the human vision. i'll make a test, which leads to better results after dot removal.
13 had center af points (and some additional edge detection)
the 2 remaining (20 and 84) do have edge detection dots...

i still don't understand why a skipped frame should move the crop position. and the offset is also not completely centered without any skipped frames. i'll check if it's always just 2 pixels more to the right, or if it varies from resolution to resolution.
as soon as i have a working version for all resolutions i'll upload it.

@haomuch:
you find nearly all versions at https://bitbucket.org/OtherOnePercent/tragic-lantern-6d/downloads/
just the version from 22. june (which is imho best so far) isn't there... but you can find that at:
http://www.qfpost.com/file/d?g=gyGwDYIUQ

haomuch

Quote from: mixer2 on June 24, 2013, 07:23:55 PM

@haomuch:
you find nearly all versions at https://bitbucket.org/OtherOnePercent/tragic-lantern-6d/downloads/
just the version from 22. june (which is imho best so far) isn't there... but you can find that at:
http://www.qfpost.com/file/d?g=gyGwDYIUQ

Thanks for that links very much. But as I found out they are like patches instead of a whole fireware file. Could you tell me how can I use them?

mixer2

Do you have ML (from the first post in this thread) already installed?
Then just extract the ZIP of the update and copy the autoexec.bin to the root of your sd-card (overwrite the existing one).
Then copy the MODULES directory into the ML directory on your sd-card. If there is not a "SETTINGS" directory in your ML directory you may create one (you don't have to, but it will show a warning if it isn't there).
Turn your camera on (in movie mode) tap with both fingers on the screen to get into ml menu. switch to the "M" tab (last tab).  select "Load modules now..." you should see some green dots in front of each module listed below after that.
switch to video tab and activate raw video. there are some things you should know:
1. you can use fps override to record less than 30fps and get continuous recording on higher resolutions
2. 1600x542 is imho highest resolution for continuous recording
3. you can switch to photo mode and activate multishot nr to get larger buffer (more recording time with resolutions that are to high for continuous recording). if you do that you've to use menu key to start recording. set first multishot nr and then activate raw video and configure the options to use menu key to start recording. and you may have to use expo override to get correct exposure (i think you still have to).

haomuch

Thanks for that explanation, very clear now

1%

P.S. all creative filters crash for me... but why in the world would you use these... I think actually need to find the regs they use and have them real time for H264 if possible.

mixer2

1%: The camera model header information in the dng isn't correct. EOS-M says "Canikon" for both, the silent dng and the dng converted from raw video. 650d says "Canikon" for dng converted from raw video and the correct model for silent dngs.
Can you check if it's possible to get always the correct model name?

1%

Will have to look... plus all our raw types seem to have the dots.. I looked through the 100 DNG and everything was F'd.

funkysound

Hi - thanks for working hard on the EOS M which acts very good together with my 600d. :D
Will there be audio funktions too, like external/internal mic and audio out for example? Do you have any idea when the next official ML Version incl EOS M will be available?
EOS 600d, Tamron 17-50mm 2,8 with stab., Tamron 70-300mm, Canon 50mm 1:1,8, Sigma 10-20mm 4-5,6, Sigma 30mm 1,4,  EOS M with 18-55mm, 22mm and EF adapter, tons of more glases incl. FD/FL, EOS 70D, Canon 18-135mm STM

1%

Beeping/wav caused lockups... dunno about other functions yet, should have the same chip as 650D but not reading/writing registers correctly yet. At the least hope to find monitoring.