Menu

Show posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Show posts Menu

Topics - mkrjf

#1
Would the 1D-C support a ML like hack?
If new 5DMk4 with video enhancements comes out, would it likely support hack?
#2
General Help Q&A / 5DMk3 1.3.3 firmware
March 02, 2015, 05:02:56 AM
My 5DMk3 was serviced at Canon Professional Services two weeks ago and several circuit boards were replaced to correct the camera not always powering up when switched on.
They did not provide details, but the work order stated they replaced several circuit boards.
Tried to run new version of ML targeted at 1_3_3 (thanks Chris!!) but apparently the boot flag that used to be set in my camera was reset (or maybe the circuit board with eeprom? storing the flag was replaced). The techs did not mention anything about ML and of course I removed the ML SD card prior to service.
So the current boot flag setting program will not run since the camera detects it is older than 1.3.*

Am I the only one in this situation? Forum search for 1.3.3 did not return anything.

It has been suggested that if I manually downgrade then flip the bit with boot flag program and then upgrade it will preserve the flag state and I will be good to go.
Can anyone confirm this? And couldn't the version embedded in the .fir be changed to be 1.3.* even if done with a hex editor on the binary?
Thanks,
Mike
#3
The following has before and after of 5DMk3 raw.
Used older version of ML that outputs raw and then converted with new raw2dng that support prores4444.
When filming I made effort to not overexpose, but prores44444 initially appeared overexposed - guess that is just how raw2dng converts.
Is it intended to output linear? I was expecting log - maybe we could have an option for that?

Just imported prores4444 into Speedgrade and adjusted exposure, saturation, black levels quickly to eliminate hot spots on white costumes.
So there was at least 10bits of information captured!
Used linear and 5DMk3 profile settings in speedgrade.
Curious to know if others have speedgrade settings recommendations or custom profiles.
Will also try with davinci.

https://plus.google.com/photos/109692674038873146393/albums
#4
Raw Video / Raw goals for cinematography
September 10, 2013, 09:53:25 PM
It seems like the direction of magic lantern raw is headed off track?
Is the goal cinematography or programming adventures?
If the former - then a stable 24fps HD raw with workflow that does not reduce quality should be primary objective. If you expose correctly with 14bits of dynamic range that is enough - the HDR and ISO bit tweaking needs to be motivated by quality issues. HDR requires high frame rate also.
Likewise for changes in raw format / in camera processing (demosaic).

My personal experience using June build on 5dmk3 is that there is more dynamic range seen by camera than makes it through workflow. Using ml member provided utilities (thx!) to get DNG and mov in one step - it sure seems like much less than 14 bits of data is making it to DNG and then mov is only prores 422. Maybe a test raw 'ramp' can be created to verify?
Is it not possible to carry 16 bits (14 data) to DNG and then full 4444 to prores?
422 is better than h264 in camera but not enough to use for indie feature work.

I have asked a couple times (no real answers) on actual processing of raw and actual bits of info in raw.
Based on ISO discussion it seems that there may be a few bits of dither 'noise' in the data. Can someone summarize that in one - two sentences?

I did a sunset test with June build with sun just over exposed and I was able to do light color grade and get a very nice crushed black and very natural looking highlight and midtone. I did not see a lot of noise in the black - but then I exposed properly and crushed the blacks a little (which will blend / eliminate minor noise in black).
So it seems the June build was close to usable - just needed some cleanup / documentation.
So what exactly is the goal of this development group?

Also I believe the frame rate is off and there appear to be some dropped frame effects even though the recording is continuous.
I will do sync sound recording (zoom or tascam) with head slate and tail slate to confirm (audio duration versus video duration between head and tail slate).

I think I would like a branch of the code that is purely targeted at indie cinematography.
Thx
Mike
#5
Feature Requests / one menu setting for cinema node
June 30, 2013, 10:15:49 PM
Take all the miscellaneous settings that contribute to good best cinematography and make one click menu item
So say 1920x1080 center cut (true center) at 23.976 fps with audio (if possible - even if low bitrate just for sync track)
Histogram displayed until rec pressed
#6
I have been struggling with getting CF cards to record continuously in raw mode on my 5DMK3
I posted a request for details on how the sensor is read and processed before writing to file NO RESPONSE :(
I suggested that maybe the camera circuitry has different capabilities by camera serial # / build date NO RESPONSE :(
But in trying all options I finally got around to trying the '5x crop mode' - actually 3x I think where a 1:1 1920x1080 portion of sensor is extracted, processed and written to CF card.

Guess what? this mode writes to the card as much as 20% faster. So how the F can the card be the bottleneck?

More likely the ML processing is the bottleneck and speed of processing depends on circuitry in the camera.
Not unusual for there to be variation as long as they all pass minimum engineering tolerance. So probably my camera exceeded minimum engineering spec less than some others. Would be nice to know if there is a pattern - as then we could look for cameras that have the higher speed circuitry.

But main point is - it would not be a lot of effort for senior developers to create a minimum flow chart showing how the two modes use internal code / hardware. And it is not god science to post charts showing absolute write capabilities of cards when it is actually showing build and camera dependent write capabilities. IMHO

So for now I will work with crop mode (and its quirks ;).
Maybe the code for 'full sensor read' raw can be optimized (or at least measured) to make it work across more of the 5DMK3 cameras.

And I REALLY appreciate the effort of this community - just looking for a little more communication and science.
Mike
#7
I didn't see a section in FAQ or forum about ideal settings for a particular card and camera.
Nice there is a spreadsheet of what is possible - but how you configured to get there is even more important ;)

I purchased transcend 1000x 64GB since several people say that works for continuous 1920x1280 at 24fps.
$400 later - no it does not work for me :(
Followed what guidelines are available - but some information appears to be missing.

Running 5 min benchmark it lists 82MBps with 16384 buffer - and 80MBps is listed as requirement for 1920x1280 24fps continuous. So is the buffer used by ML smaller than the ideal buffer measured in benchmark? I don't see actual buffer listed or a way to adjust the buffer. There is a read only value in the ml menu - but it does not appear to be the same.

If I don't touch fps override camera captures 18 frames and stops recording. (Same for 400x 32GB card!). If I set fps override to 23.976 toggle - then 45 frames are captured before automagically quitting for 1000x 64GB
CF card.
Using:
June 10th (ed914b3).
Any ideas? Are results very different across same camera model?
Mike
#8
I will try to consolidate all my questions on this thread - the focus is maximum quality for indie films with cinema quality post production.
Workflow (super high level):
1) Shoot in camera (5dmk3 with L series glass and neutral profile) with ML at 1920x1080x24fps and approx. 1/48 shutter. Use histograms to primarily avoid any overexposure and secondarily minimize underexposure. Obviously you will adjust exposure, aperture, lighting, lenses, etc.
2) Ideally edit with system that can use 'proxy' for real time editing of raw files but debayering / transcoding to an almost lossless coded may be good enough if exposed well - so maybe premiere against raw / tif.
Convert raw files generated to a lossless intermediate codec for coarse color correction and subsequent editing - say prores444
In either case the bit depth must be preserved as much as possible - silly to shoot 14bit just to truncate to 8bit at the start of workflow.
3) once you have picture lock - either export at highest level (again could be prores444) and grade directly or export metadata that allows grading against raw files using metadata to delineate edit decision list / xml.
4) Now after you have done all your magic - consider compressing / downrezing for a particular output - say web or tv or whatever.

If everyone agrees with this - then can we specify what is the maximum quality that can be output with ML (straight - no HDR or other tweaks for now)? What is a workflow that is not restricted to one commercial tool (ie works for fcp and premiere and whatever - does not hinge on after effects) that generates this best case intermediate codec for editing.

If that can be done you will cause major disruption and everyone will benefit.
IMHO

For quality please consider spatial and color and time - line pairs / mm, pixel merging algorithms (when converting full sensor 9:1), debayering algrithms (still haven't seen any details in this forum on algorithms), color space tranformation, exposure angle / duration, etc.

Mike
#9
General Help Q&A / 5D Mk iii MLtechnical specification
September 21, 2012, 05:12:48 PM
Before I take the risk of bricking my new camera, can you clearly list what the software capabilities are (for that specific camera and ML release)?
I am looking for 4.2.2 output for I frame at 1080p24 and 720p 60 with highest bit rare and also HDR video and zoom preview.
Are any of these supported in ML 5D Mk iii alpha/beta?
Thanks,
Mike