5D2 RAW video Builds 14-Bit

[a.d.]

I've updated the vanilla build with a1ex fixed for further testing:
6dfbee990661_2nd.zip

[Africashot]

perhaps then, mu question may be off topic, but then, how does a smaller sensor, like the 50d, get more latitude, as well as a higher bitrate?

I guess, although the sensor is smaller, it is made up by bigger (and less) pixels resulting in less lineskipping, not sure about the bitrate but that is probably somehow related to the CF portal bottleneck

[Petruk]

I've got a 5d M2 and edit on a mac computer. I can't find the information I'm looking for on this thread, and I wonder if anyone who has made ML work with this configuration would be willing to email me and later answer some questions on the phone. I live in the USA and can pay you something for your trouble via paypal. email address is dandmcguire at  gmail.com.
Thanks,
Dan McGuire

[Clemens]

Somehow the latest build with this new saving method of buffer is not so much speeding up my recordings.

[bnvm]

@a.d. I have found some possible bugs. I don't know if anyone else has this but in the latest daily build, and the one before it at least, lists 1856 as the resolution rather than 1872 with 1880 turned off. It always shows as 1856 in the menu but when I select that resolution with 1880 on the resolution display on the live view shows 1880. When I disable 1880 the live view displays this as 1856. The videos are recoreded at 1856 so there is no way of getting 1872.

Also the needed MB/S indicator is off by a factor of 10, its displaying 616.6 mb/s rather than 61.7 mb/s for example.

[a.d.]

@a1lex
The fixed works

@bnvm
It's not a bug, you need to choose 1920 to get 1872

@Petruk
just post your post-processing problem in this section:
http://www.magiclantern.fm/forum/index.php?board=14.0

[bnvm]

@a.d. Got it, just a thought but it may be good to make the 1880/1872 their own independent resolutions separate from 1920, that way someone could shoot 1872 regular and in 5x crop mode. Right now it is only possible to shoot 1872 regular and 1920 crop. Thanks and awesome work.

[Will@Elevate]

I can't manage to get higher than 58mbs write speed using my 32gb Lexar 1000x on my 5d2.

I have the instructed camera settings, and global draw turned off.. Am I doing something wrong, or do I have a bad card?

Thanks,

Will.

[Doyle4]

I can't manage to get higher than 58mbs write speed using my 32gb Lexar 1000x on my 5d2.

I have the instructed camera settings, and global draw turned off.. Am I doing something wrong, or do I have a bad card?

Thanks,

Will.

As far as im aware the 5D Mkii can only write upto 60mb? i could be wrong, the 1000X speed is read speed, usually on 90mb cards it's referring to read speed.

[noix222]

I can't manage to get higher than 58mbs write speed using my 32gb Lexar 1000x on my 5d2.

I have the instructed camera settings, and global draw turned off.. Am I doing something wrong, or do I have a bad card?

Thanks,

Will.

are you setting up your camera in JPEG??

[Hazer]

I have the Lexar and it writes high 60s, occasionally 70.  As has been noted, set the camera to jpg *stills* and turn off raw stills.  Turn off ML global draw.  And finally, turn off all on-screen Canon graphics.  i.e. Press the 'info' button until the display shows only the LiveView image, not the aperture, shutter, ISO, EV, or anything else.

[Will@Elevate]

I have the Lexar and it writes high 60s, occasionally 70.  As has been noted, set the camera to jpg *stills* and turn off raw stills.  Turn off ML global draw.  And finally, turn off all on-screen Canon graphics.  i.e. Press the 'info' button until the display shows only the LiveView image, not the aperture, shutter, ISO, EV, or anything else.

I have done all these.  when you say high 60's are you referring to the benchmark results (stills/movie mode), or what you see while recording raw video?

Thanks.

EDIT: 70mb/s is the max write speed for the 5D2 correct?

[ted ramasola]

I have done all these.  when you say high 60's are you referring to the benchmark results (stills/movie mode), or what you see while recording raw video?

Thanks.

EDIT: 70mb/s is the max write speed for the 5D2 correct?

benchmarks aside, can you record 1728 x 972 continuously?

[ted ramasola]

I've updated the vanilla build with a1ex fixed for further testing:
6dfbee990661_2nd.zip

@a.d. & a1ex

Here is my comparative test between a.d. jun 25 nb "pink fix" (d07dab0ba6c2)
and that of the nb dated jun 26 "combining" a1ex "vanilla" and a.d. (6dfbee990661_2nd)

PINK cast in 5x crop modes are gone.

BUT apparently a.d. nb jun 25 seem to have slightly more frame count than that of nb jun 26.

For these tests, I want to activate features that would be useful for "serious testing".

GD= ON, Preview= Auto. FPS is still set to 23.976.

I made 2 consecutive recordings per resolution to get average # of frames except for 1856x928
were I made 4 tests each due to the high difference, i wanted to confirm the frame count.

test results are separated by ( / ).

RESOLUTION  -- JUN 25     --    JUN 26

1x MODES

1872 X 1054 -- 395/501     --    447/379
1872 X 1012 -- 882/822    --    769/710
1872 X 936  -- 2389/3359   --    2764/2011
1872 X 850  -- continuous    --   continuous
1856 X 1044 -- 612/573     --    611/567
1856 X 928  -- 8675/2445  --    4891/4267
                    /9101/9927      /2099/4891
1728 X 972   -- continuous    --   continuous
1472 X 1104 -- continuous    --   continuous

CROP MODES 3X (5X MAGNIFY)
   
2144 X 1076 -- 234/237     --    219/197
2144 X 1072 -- 211/241     --    153/178
2144 X 972  -- 353/373      --    326/302
1920 X 1076 -- 402/411     --    347/313
1856 X 1044 -- 677/794     --    611/567
1728 X 972  -- continuous    --    contnuous

Also, other features I like in the 25th NB are;

The Black border crop from pravdomil.
Though minor, the options of 1880.

Though I know this is still testing phase, the buffer meter and
other info are smaller and out of the preview frame.

About the preview during recording in crop mode, is it
possible to have the option to choose between, ML grayscale and the
one in 25th build where you get realtime view and color? in the 26th NB
only ML grayscale is there. its ok since framing is accurate but preview
lags so much.

[a1ex]

Did you have the same overlays enabled while recording, in both builds?

I did not remove the Canon preview during crop mode. It's grayscale while recording because of image quality issues with the color preview (I've explained it a lot of times).

[ted ramasola]

Did you have the same overlays enabled while recording, in both builds?

I did not remove the Canon preview during crop mode. It's grayscale while recording because of image quality issues with the color preview (I've explained it a lot of times).

yes. Both have the same settings. cards are formatted everytime I change builds and also before i do the long recordings.

I understood your concerns on the preview.

[a1ex]

I've checked the code and can't explain the speed difference. Can you compare the latest vanilla (#625) with the one that still had pink cast issues? (#597)

I've noticed the overlays are turned off while recording in a.d. version, while on vanilla builds you have to do so manually.

Also check whether the predicted frame number is achieved. That one is based on the ideal mathematical model: buffer_size / (capture_speed - measured_write_speed). If you are reaching that number, the buffering strategy is optimal (you can't get any better than that without increasing write speeds).

If you are not reaching the predicted number (say it predicts 2000 frames and you only get 1500), we need to enable the debug graphs to see what's going on. Most likely, ML is overestimating card write speed a little.

In my own tests, the prediction was pretty good.

[guentergunter]

First of all: I tested a1ex build and was surprised how few settings there are availlable (until now I only used a.d.'s builds). That's generally great, except one setting: '1880 or 1872'.

On the first hand, 8 pixels more (and 2 the other dimension) are not really something you need to fight for.
But on the other hand, why should I drop them if they are availlable anyway (thus even crop my image a little bit (the recording frame is obviously smaller with the lower resolution)) when there's no decreasing in speed performance?

By the way: Since I joined the raw builds (about two weeks ago), I tried every new build and never had any bad pixels or anything like that. But I never used the 5x or 10x crop modes for I can't see much sense in that (for me).

[Ethan_Reitz]

Hello! Just downloaded 282abfa87fef.zip, installed and took my 5d mark ii out to a field to test for 20 min. at sunset. Unfortunately I forgot my release plate so it's all handheld. The card is a 32 gb lexar 1000. I've used ML for a while now and I didn't take too much time to read up on the new raw features - just looking at the incredible results around the web!

So here's my rough test:

So I noticed a couple of things - one clip was pink all the way through - it was at 10x. There are also some clips which stutter quite extremely - I was wondering if it was because I had my image stabilizer on and the camera wasn't moving enough? Thoughts on that? There is also a line at the bottom through a couple of the clips.

But I shot directly into the setting sun and there is visible foreground!!! Amazing! (I am using a Light Craft variable nd filter)

I used raw2dng and graded in AE - then exported to h264 720p for vimeo.

Below links to
0 is pink,  21 stutters and 18 has the line at the bottom.
https://dl.dropboxusercontent.com/u/79843444/000000.dng
https://dl.dropboxusercontent.com/u/79843444/000021.dng
https://dl.dropboxusercontent.com/u/79843444/000018.dng

Thank you for what you're doing here! Keep it up!

[guentergunter]

I've noticed the overlays are turned off while recording in a.d. version...

Quite useful

[reddeercity]

I have done all these.  when you say high 60's are you referring to the benchmark results (stills/movie mode), or what you see while recording raw video?

Thanks.

EDIT: 70mb/s is the max write speed for the 5D2 correct?

Hi,
 you should be getting at least 65MB/s- 70MB/s write to the card on  Lexar 1000x 32GB card
As of June 22nd built these are the Number i'm getting 
Test out my Scandisk 600x 90MB/s 32 GB
1880x800 24p @ 57.3-60.1 MB/s Continuous Recording

On my Lexar 1000x 32 GB , with the June 22 Built
1856x928 24p @ 68.7-69.1MB/s with frame over ride on
1880x854 24p @67.2MB/s with frame over ride on
Crop mode : 2048x856 24p Continuous Recording

In my setting i have Frame over ride On @ "24p" exact, Not 23.98
and i set my Photos to Small Jpeg.

i have Problems recording anythink over 70 MB/s

[a1ex]

@guentergunter: I have explained the reasons for mod16 resolutions, and I'll stick with that.

If a.d. wants to provide 1880, that's his own choice and I do not support it in any way. I know what trouble I had on 5D3 with random lockups because of DMA misalignment, so I'll stick with a portable solution that works everywhere.

As for the other options, some of them are good and will be added to vanilla builds (e.g. black bars, but in a clean way that does not fight with other overlays), others are just ugly hacks (e.g. dropping frames instead of optimizing playback code, or some status displays moved over ML overlays, therefore flickering like crazy if you enable both.

[ted ramasola]

I've checked the code and can't explain the speed difference. Can you compare the latest vanilla (#625) with the one that still had pink cast issues? (#597)

I've noticed the overlays are turned off while recording in a.d. version, while on vanilla builds you have to do so manually.

Also check whether the predicted frame number is achieved. That one is based on the ideal mathematical model: buffer_size / (capture_speed - measured_write_speed). If you are reaching that number, the buffering strategy is optimal (you can't get any better than that without increasing write speeds).

If you are not reaching the predicted number (say it predicts 2000 frames and you only get 1500), we need to enable the debug graphs to see what's going on. Most likely, ML is overestimating card write speed a little.

In my own tests, the prediction was pretty good.

I have data from tests with vanilla, give me a couple of minutes while I run some confirmatory tests.

[ted ramasola]

Comparing Vanilla (5d2-fac94f686d23-pink-debug ) and a.d. pink fix jun 25 (d07dab0ba6c2)

Ran tests 2 times consecutive for each resolution except in 1856 x 928, ran it 4 times



Resolution     --  A1ex             --  a.d. jun 25 pink fix
                      Vanilla jun 25
1X

1872 x 1054  --  390/528       --  395/501
1872 x 1012  --  840/709       --  882/822
1872 x 936    --  3667/2009   --  2389/3359
1872 x 850    --  continuous   --  continuous
1856 x 1044  --  510/436       --  612/573
1856 x 928    --  3989/5177   --  8675/2445
                            /3321/6452      /9101/9927

1728 x 972    --  continuous   --  continuous
1472 x 1104  --  continuous   --  continuous
   
3X Crop mode
   
2144 x 1076  --  188/235       --  234/237
2144 x 1072  --   221/218      --  211/241
2144 x 974    --   487/376      --  353/373
1920 x 1076  --   416/386      --  402/411
1856 x 1044  --   736/748      --  677/794
1728 x  972   --  continuous   --  continuous

Results are close except for that in 1856 x 928

[guentergunter]

@guentergunter: I have explained the reasons for mod16 resolutions, and I'll stick with that.

If a.d. wants to provide 1880, that's his own choice and I do not support it in any way. I know what trouble I had on 5D3 with random lockups because of DMA misalignment, so I'll stick with a portable solution that works everywhere.

As for the other options, some of them are good and will be added to vanilla builds (e.g. black bars, but in a clean way that does not fight with other overlays), others are just ugly hacks (e.g. dropping frames instead of optimizing playback code, or some status displays moved over ML overlays, therefore flickering like crazy if you enable both.

It's a theme with two sides...
On the one hand, it's about in-camera write speed. That's something I absolutely trust you (since I don't know much about those algorithms), when you say it decreases write performance (although I do not suffer from that with 1880).

On the other hand it's about h.264. Let's suppose someone owns a 5D2 and does not want to crop the image (5x or 10x) due to the loss of wide angle perspective. So his highest possible resolution is 1880/1872. But noone will use these resolutions for anything further. Thus it's resized to (most obviously) 1080p or 2K.
By scaling an image that way, any previous mod16 resolution will become absolutely needless, unless the primary resolution overall is as high as possible. And here he goes to 1880.