Tragic Lantern for EOS M

[haomuch]

I don't see much meaning in crop mode video. The crop result of a 22mm lens is definitely far different from a full 50mm f/1.4 lens effect. Especially in terms of field depth.

[mixer2]

@JohnBarlow:
i'm uploading some raw files for you at the moment. 3gigs... it may take some time. as soon as it's ready there are 11 raw files. resolutions 1280x720, 1024x768 and 1088x816.
https://www.dropbox.com/sh/ofaqeiwwjg0xdrw/HAYYgd9cXk

there are also some h.264 shots with 18mm crop and 55mm non crop for comparison and a 1280x720 24p webm file with the processed raw video files. no real post processing on the webm... just used the default canon raw curve to get some contrast into the image, no noise reduction or sharpening.
http://dl.dropbox.com/s/k8fe71ga1kiw57o/720.webm

720p raw looks much better than i thought. and so much better than h.264 1080p, even without processing Oo
but of course i could have done h.264 better with reduced saturation, contrast and sharpening... it's auto picture style.

can anyone check if faster card really makes a difference to get continuously 1280x720 24p recording?

[JohnBarlow]

@mixer2

Thanks for your time, it has convinced me to get an EOSM with MLRAW,
 
who needs a Black Magic pocket cam?

[mixer2]

anyone has experience with the SanDisk SDSDQX-064G-U46A... it's a microSD, but benchmarks look nice and price is very moderate.
i really want to know if we can get more speed (maybe really 36mb/s continuously) with a faster sd-card than sandisk extreme 45mb/s.

or maybe anyone has the 45mb/s sandisk extreme and the 95mb/s and can compare them directly in the eos m.

[gravitatemediagroup]

Anybody know anything about this clip other than what's mentioned? Like exact ML settings and workflow?

[mixer2]

it was posted by qsara on page 19 of this thread. but the qsara account seems to be deleted.

as the video says it's 1664x464 30fps non crop mode. you can't get that resolution anymore, but 1664x456, if you don't need continuous recording at 24 fps, or 1664x428 (1600x438, if you prefer that aspect ratio) for continuous recording. as the video is recorded at 30fps i think the 21secs was the maximum that could be recorded on that resolution and framerate.

as qsara said on page 19 it's recorded at 100iso. there are not much more camera parameters... i guess medium focal length and aperature wide open. there is no way see what shutter speed is used without motion in the video, but with 30fps normally 1/50 or 1/64 is used.

in post it got unsqeezed and some color correction (i guess whitebalance, added saturation, added contrast and it looks like a warmify filter or something like that was added). since there was no dot remover at that time and it's hard to spot any dots, i think the dots got removed with a dead pixel filter in the raw editor.

everything just guesses, don't really know what qsara did. but if you tell us what exactly from the video you try to achieve maybe anyone finds a solution.

[markr041]

I made a video from shooting RAW at 1280 x 720, 24 fps, in crop mode. Using a Sandisk Extreme sd card, I was getting about 35 MBps and could shoot, according to the camera, over 2400 frames before stopping.

My workflow:

Dot-removed the RAW clips.
Obtained the .dng frames from the dot-removed RAW files.
Used Lightroom 5 to convert the .dng files to uncompressed .tiff files using default settings.
Used Vegas Pro 11 to edit the string of .tiff files to create the 720p video at 24p, using AVC compression and light sharpening using unsharp mask.

[Rush]

Is this the same movie crop mode in ML? I still get moire with this on. Have you tried shooting a test chart?

I can get it, but it is very very little, comparing to non-crop mode.

technically 100% crop doesn't mean no aliasing. i read that multiple times, but it's still a digital measurement, which means, if you're signal (the light that comes from the lense) has more detail than your sensor resolution you'll get aliasing effects.

technically 100% crop don't use lineskipping. aliasing and moire comes from the gaps of lineskipped pixels, and gaps becomes minimal in 100% crop mode.

[mixer2]

@markr041:
looks nice. it would be interessting if you could get more frames.. as i wrote it stopped for me after 2897 frames...

@Rush:
"aliasing and moire comes from the gaps of lineskipped pixels"

afaik, that's not true... any references?
it's all about frequencies... if you've too high frequencies (not the frequency of the light, but the singal frequency) for your sensor, then you'll get aliasing. check the "Nyquist theorem". http://redwood.berkeley.edu/bruno/npb261/aliasing.pdf
it's a gernal problem of sampling analog singals.
of course line skipping makes it even worse, because it heavily reduces the sampling frequency.
the only reason why you don't see much aliasing at 100% is that the lenses aren't that sharp and act like a low-pass filter.
=> sharper lense = more aliasing

[Rush]

of course line skipping makes it even worse, because it heavily reduces the sampling frequency.
the only reason why you don't see much aliasing at 100% is that the lenses aren't that sharp and act like a low-pass filter.
=> sharper lense = more aliasing

yes, I agree. But for me - without lineskipping aliasing is acceptable and hardly noticable (even with the sharpest lens I have), comparing to lineskipped image where aliasing become obvious.

[mixer2]

@1%:
since dual iso does also work on 6d and 5d ii, is there any chance that it may also come to eos m? or does it just has a single amplifier?

@JohnBarlow:
the large upload of the test raw files from yesterday is now finished.

[JohnBarlow]

@mixer2

Got them, Thank you very much.

Can you confirm my current understanding?

Crop mode is about 3x, giving a recording sensor width of 22.5/3 = 7.5mm

720p would give about 2/3 of that resulting in a recording sensor width of 5mm

Is my calculation correct? Thanks

[mixer2]

EOS m has
5184x3456 pixels on 22,2mmx14,7mm

with 1280x720 crop mode you use 5184/1280 ~ 24,69% of the horizontal and 720/3456 ~ 20,83% of vertical pixels.
14,7mm*0,2083 ~ 3,06mm
22,2mm*0,2469 ~ 5,48mm

sqrt((3,06mm)²+(5,48mm)²) ~ 6,27mm

if my calculations are correct you'll use a diagonal of ~ 6,27mm with 720p

43.2mm/6,27mm ~ 6,89 <- this should be the effective crop factor

compared to bmpcc which has a super 16mm diagonal of 14,54mm and 2.97 crop factor, it's a really tiny sensor. less than 1/4 of effective sensor size.

[JohnBarlow]

Thank you for a very precise answer

[manly]

Here is my understanding of the problem. Feel free to correct me.

The only real limitation we have is how to fit a 16:9 aspect ratio in a 3:2 sensor. In other words, 5184 pixels wide (or rather, a bayer-patterned mosaic of 5184 pixels wide) by 2916 pixels, cropping 540 pixels top+bottom combined.

Our cropped 5184x2916 (16:9) video is clearly above 1080p (and the write speed). We only need to resize those into either 720p (1280x720) or 1080p (1920x1080). If the digic5 processor is having a hard time doing this, then how about, in the case of 720p (where we need to cut to 24.69%), take 8x8 pixels (bayer-pattern) and reduce that to a 2x2 by averaging the bit-depths of the corresponding colors to the smaller matrix? You would get an exact 1280x720 pixel (720p) by cropping to 5120x2880, which would be fairly close to the full 5184x3456. This operation, under x86 cpus, would be both easy to code and quick to execute.

For 1080p it would be a bit trickier, since 1920/5184 = 37.03% doesn't rounds nicely. I'll let the knowledgeable people first see if my idea makes sense then I'll see if 1080p raw is even feasible (might not even be with write speed bottleneck).

[mixer2]

@manly:

sorry, no... i think everyone wishes that would be the "only real limitation".
it's not possible to record 5120x2880 because of the speed of the sd-controller. there are 2 raw streams to choose from. one is downscaled via row and lineskipping (which leads to really bad quality) and the other is a 100% sensor crop (1792x1028 cropped from the center).
but, since 1792x1028 are still to much pixels to record (continuously... you can record a few frames till it stops), it's even more cropped to ~5% of the sensor size.
=> you can get thise 5% of the pixels from all over the sensor with a lot of aliasing, or from the center, which leads to a high crop factor.

main limitations are sd-controller speed and the bad quality of internal downscaling.

[handbanana]

@mixer2:

I'm still puzzled as to why the non-crop mode is squished. That doesn't happen on my 60D or T2i, although I did see there was a squished option when recording at 720p resolution in movie mode on the 60D.

[mixer2]

i think the eos m handles the downscaling different from other canon cameras... but really don't know.
have you watched the video i posted some days ago: http://dl.dropboxusercontent.com/s/p1qo960nrhmcd7g/crop_vs_non_crop.webm

the first 10 secounds show crop mode vs non crop mode... and what i see is that non crop mode has really bad quality compared to crop mode.
maybe 1% can tell you why the downscaled videos on other cameras don't look that bad.

[Eran]

It will be interesting to see a comparison with the BMPCC Prores in the near future . I wander if anyone thinking the same as I do.

[feureau]

It will be interesting to see a comparison with the BMPCC Prores in the near future . I wander if anyone thinking the same as I do.

Speaking of prores. I wonder how they managed to put that in the BMPCC when the canons don't have enough computing power to convert raw to prores... Dedicated prores chip?

[haomuch]

Hi guys, any news about the magic zoom function improvement? I heard a new patch was just released.

[1%]

MZ is at least in sync in most modes, but not much different.

Its ugly at idle because its always 720P sized. Other cameras ugly mode is only 60FPS and all modes are like crop mode minus the crop.

[haomuch]

MZ is at least in sync in most modes, but not much different.

Its ugly at idle because its always 720P sized. Other cameras ugly mode is only 60FPS and all modes are like crop mode minus the crop.

Err, not quite understand the statement. Just want to know if it's possible to use the magic zoom with higher resolution according to the magnification factor?

[handbanana]

@Mixer2:

Yeah I saw the video, and even based on further testing I did I've seen what you mean. Plus upscaling that much is still not ideal. But what I mean is on the other cameras the raw video in regular (non crop) mode isn't vertically squished so when you upscale you upscale both x and y.

[Drizzt321]

Speaking of prores. I wonder how they managed to put that in the BMPCC when the canons don't have enough computing power to convert raw to prores... Dedicated prores chip?

Either implemented in hardware, or more likely using some GPU Compute resources to do some of the work (the latest Tegra based design comes to mind) or a dedicated DSP plus a multi-core ARM chip probably makes it doable without being too expensive either from a cost or power design. It'd be interesting to know what they've done actually. Although, I don't actually know that ProRes is really a heavily compressed for mat. Probably doesn't take much more computing resources over h.264, just a higher write bandwidth is required. Maybe more CPU if there's no hardware dedicated to it.