Danne's crop_rec_4k experiments for EOS M

[henricusmaria]

Thanks for feedback guys.

@henricusmaria
12bit is better and very close to 14bit in quality. 10bit will work most of the time imo.

Sorry to come back at this, I think I wasn't clear in my question. I wanted to ask if there is a big difference in using the 10, 12 and 14 bit uncompressed raw settings. Or is it better to shoot in lossless 14 bit raw? I've been doing some footage in movie crop mode with the 12 and 14 bit uncompressed raw setting. Is the difference between uncompressed and lossless neglible? Thanks

[Danne]

Thanks Upstep for tests. Those are my exact observations also. I did some tweaks further very late yesterday. If you or others could run this build and it's not worse or maybe better I will say I keep this version.

@henricus....
There is no visible difference between lossless and uncompressed. Test, compare, read and you'll see why.

Regarding bad frames. The cpu runs much heavier when framing is running. Realtime view not so very much. I guess if one stays longer in real time modes chances for less bad frames are bigger.

Also. It is possible that the sd_uhs overclocking procedure is causing trouble. I never had a corrupted frame when not using overclocking with 1x3 binning modes.
What can I say. It's bleeding edge.

[zzzeosm]

Are there any instruction for new builds？ i was got some troulble with the setting menu.
thanks for all you guys doing and testing with eosm.

[masc]

I did some tweaks further very late yesterday. If you or others could run this build and it's not worse or maybe better I will say I keep this version.

I am not sure if it is because of much light outside today, but in 20sec I got 5 corrupted frames with latest build. So it seems not to be better than the build before for me.

Edit: in same conditions as yesterday (office room) I got for another 20 sec clip 0 corrupted frames... hmmmmmm

[AF-OFF]

hi
just tested Danne's new Jan18's build with 1x3 preset.
1504x1920, 24fps, with audio.
MLV.app on HS 13.1 - now exporting as 4512x1920,
no single corrupted frame. recording continuous, the longest clip ca. 45sec.
looks promising, will test also the HD and 2,5k presets later.

really awesome work Danne!

[Danne]

Great! Been testing timing registers all day. The last three uploads seems to work really well.
Also make sure you tested the latest. They are all named the same today .

[bergamote]

Man, I can't keep up with this frenetic development schedule 
I tested the 1x3 mode in dual iso with the 14jan build (reliably corrupted frames), then the 16jan build and had no corrupted frames but had to be quite underexposed for it to be continuous. On ML's histogram I have to leave between two and three columns (EV?) on the right empty.

About dual iso processing in MLV App, is there a sure fire way to bring the shadows back, like say +1 exposure or +X shadows, or you just play it by eye?
Also, sometimes I do get some kind of flickering in the shadow/midtones, exactly like @loknar has in some shots of his windy winter day vid (great footage btw, shame youtube compresses the hell out of it. Maybe uploading in 4k...?).

Gonna try today's build tomorrow (by which time it will be totally outdated  )

[bobolee]

66 yuan（about 10 US dollar）for the fake battery kit，able to take hold of np-f970 f770 f550 etc，50 yuan for the grip，128 yuan battery kit：third party np-f970 with a charger，sdxc：128g sandisk extreme pro
I've tried recent builds,I think the most usable mode is mv1080p 23.976fps raw on 1736*976(16：9) 14bit lossless.Need to use sd overclocking.Post processing:mlv app,convert to cinema dng,drag and drop to premiere pro cc 2017,my complaints are the noises and hot pixels,mostly moire pattens. Other modes seem to have all kinds of issues which make them purely experimental not practically usable as a proper video camera.Looking forward to have more usable mode,thank you.

[AF-OFF]

just tested again my footage from yesterday and realized I made a mistake - have to turn off the "drop frame mode" in -> MLV App/Display.
unfortunately out of nine clips, in three of them, one frame is corrupted.
will do further testing later.

many Thanks to Danne and others!

[Danne]

It´s cool. Still a work in progress:
https://www.magiclantern.fm/forum/index.php?topic=9741.msg208959#msg208959

[Jonit]

Danne, this is unreal! 
I have been playing with it yesterday and today for a couple of hours... testing high res, high fps and dual iso modes.
This is like dream coming true slowly, Its really blowing my mind what you guys can do with these tiny cameras.
Especially I am blown away by 50fps mode. I set fps to 48fps and horizontal res to 1600px (2.35:1) and it is working superbly. Now and then it skips or freezes on one frame but other than that, it is awesome. I think my card is not fast enough to sustain higher res, cos I am getting somewhere arounf 35MB/s writing speeds with SD hack on. (SanDisk Extreme UHS-3 32GB)

In other modes (1x3 binning - the last one) I am getting a few pink frames.

Anyway.. I am super stoked about this!
Thank you and keep up the awesome work Danne.

[Danne]

Great to hear that it´s working. Almost without quirks. All reports on corrupted images are welcome. I am still tweaking and testing.
Yes, a faster card should work with around 54-55mb with the hack enabled. I am using a sandisk 95mb extreme pro I think it´s called. Not sure what other cards are working or not. A1ex has all the insights...

By the way Jonit. What 50fps setting are you using? There are two:
CROP_PRESET_3x3_mv1080_50fps_EOSM
CROP_PRESET_3x1_mv720_50fps_EOSM

[AF-OFF]

hi
just downloaded Danne's last jan19 build and did some little testing with the 1x3 mode
from 5 clips only one has had two corrupted frames just from start - the first white, the second 'normal corruption'
waiting for tomorrow to do more tests.





Image seems to look identical to what I saw on the camera display
settings - 2,35:1, realtime LV, sound, 10bit.

[ctfire]

Great to hear that it´s working. Almost without quirks. All reports on corrupted images are welcome. I am still tweaking and testing.
Yes, a faster card should work with around 54-55mb with the hack enabled. I am using a sandisk 95mb extreme pro I think it´s called. Not sure what other cards are working or not. A1ex has all the insights...

By the way Jonit. What 50fps setting are you using? There are two:
CROP_PRESET_3x3_mv1080_50fps_EOSM
CROP_PRESET_3x1_mv720_50fps_EOSM

Also have the same sandisc card but with sd_usd hack the speed not getting higher than 35mb\s ... with old builds it was around 45 mb/s

[Danne]

You can test your card through bench  module(not included in this version but get it from one nightly build).
If you run most presets continuous when set to 10bit, 2.35:1, sd_uhs hack on it should be ok. Works here.

[AF-OFF]

hi
some results testing Danne's last jan19 build.
camera settings from canon menu - videosystem NTSC, 1920x1080@24
camera settings from ML menu - Overlay>GD on all modes, Focus peak ON, Movie>RAW video>Preview Real-time (Halfshutter press to frame, works great!) , Sound ON


1.,    the mv1080p 1736x1158 mode (which is 3:2 ratio, same as the sensor, could this be fixed in the croprec/ratio menu? ) records 1736x976 (16:9), (does not change even if I set Aspect ratio to 3:2 in RAW video) @ 24fps without problems. very nice. did a 15 min. recording, cam temperature stays at 48°C, no frames corrupted. green record indicator all the time.
2.,    the mv720p 1736x696 50fps mode, longest continuous recording ca. 1min. with average of 30-40 sec. continuous, has some corrupted frames, out of a 1200 frames recording it has 14 corrupt frames, another 300 frames recording has one. the bottom of all clips has a approximately 10 pixel thin line in all clips with 50fps. all clips have to be transformed in MLV App to 1,67x height.
3.,     the mv1080p 1736x738 50fps mode, continuous recording between 500 to 800 frames, no bottom pink line, ca. 3-4 corrupt frames per clip.
4.,     the mv1080p 1736x976 45fps mode, continuous for average 200frames, the beginning of most clips heavily corrupted, then ok, one clip without any corrupt frames, no bottom pink line. no transform required
5.,     the 2.5k 2520x1418 mode, 20fps, no corrupt frames, continuous (did 5min. recording)
     
------changed my canon video setting from NTSC, 1920x1080@24 to PAL and 1280x720@50
6.,    the mv720p 1736x696 50fps mode again , max 15 sec. record time, the thin whitisch bottom line still there, the many corrupted frames at the beginning not there anymore, got ca. 5 corrupt frames out of 500
------changed my canon video setting from NTSC, 1920x1080@24 to PAL and 1920x1080@25
7.,   the mv720p 1736x696 50fps mode again, longer recording time up to 20sec., whitish bottom line present, out of 1200 frames only six or seven corrupted
------same canon settings ( PAL and 1920x1080@25) - this didn't work, changed back to NTSC and 1920x1080 30p, broke the whole thing, reseted everything, using again NTSC and 1920x1080 24p.

    got the bench.mo from the experimental lua build, skips read speed, no bitmap file saved, write speed of the sandisk extr. pro 128gb goes up to 56 MB/s, one time read speed was reported to be 72 MB/sec - this would be great as write speed .... just kiding

used the august build before, which was good but more unstable, sometimes a complete clip was corupted, once or twice I've got a corrupted frame in a 2.5k crop rec. more difficult to get it running at all, now this changed dramatically, one can switch between the modes easily.

many thanks to Danne and others!

[AF-OFF]

@ Danne >thanks for the new jan20 version!

[Tullen]

SUPER BLOOD MOON.

This development speed is insane and great. In current options, what would you recommend for capturing the super blood moon tomorrow morning? I have a 220mm F4.0 that I though I could use with the options with highest crop and best light handling. Or should I go for the 135mm F2.8? Dual iso I guess would be good too. Any suggestions for settings and workflow?

[AF-OFF]

Does anyone have a recommended setting/resolution for least jello with best resolution?
The 4k resolutions are giving me serious shutter roll if I move even the slightest.

as far I understand this, try to use a shutter speed which comes closest to 180 degrees - like 173° or 191°.

[AF-OFF]

SUPER BLOOD MOON.

This development speed is insane and great. In current options, what would you recommend for capturing the super blood moon tomorrow morning? I have a 220mm F4.0 that I though I could use with the options with highest crop and best light handling. Or should I go for the 135mm F2.8? Dual iso I guess would be good too. Any suggestions for settings and workflow?

thanks for the info Tullen. and good luck with it! hopefully there are no clouds
I would suggest to use the f4 220mm . also the 2.5k preset, and lower the framerate so You can use even 5.6 or 8 as aperture, to get the best sharpness of the lens, if this is not at f4.
dual iso maybe not that good for this.
hope to catch it here in Berlin too

[Tullen]

@ AF-OFF

Thanks for the help. Any recommendations for how to get proper exposure. Since the light falling on the moon will change as the shadow disappears.

[AF-OFF]

I would take care of the sensor saturation. like overexposing a bit rather then underexposing.
hard to tell, because the moon is still very bright, but I think there would be a better result if You recover highlights then shadows

[Tullen]

Regarding motion blur. What shutter speed do you think is good? 1/125 or faster?

[Danne]

AF-OFF. Thank you!
Will dive into this when I get more time. Goal to have solid none corrupted recordings. We'll see how far it goes...

[loknar]

Regarding motion blur. What shutter speed do you think is good? 1/125 or faster?

You can compute it.
Moon does (approximately) 360 degrees in 24 hours, that's 12 degrees in one hour, that 720 arcminutes in 1 hour, that's 12 arcminutes in one minute, that's 720 arcseconds in one minute, that's 12 arc seconds in one second.
Now you lens have Field-of-view, and you sensor specific resolution, so you can compute what angle one pixel covers and then you can compute by how many pixels FOV shifts and then you want is to use time during which moon moves less that half of pixel.