Question for 5D shooters

[Sean Cunningham]

When you are shooting with ML and defining a 4:3 aspect ratio are you using the same height as normal 16:9 shooting or does this force you into a center-cut crop mode?

[QuickHitRecord]

ML gives you several aspect ratio options, and then resolution options therein. To use as much of the sensor as possible (at least vertically), you can choose up to 1600x1200 (though I have found that continuous recording is iffy at this resolution). The smaller the resolution, the greater the crop factor. Hope this helps.

[Sean Cunningham]

Mmm, sorta, I'm mostly just wondering if you get to use the full 24mm sensor height when defining a non-16:9 aspect ratio.  I don't have access to a Canon camera at the moment to satisfy my curiosity.  I went looking up the "crop mode" once and the online ML docs weren't entirely clear to me about how these various resolutions and aspect ratios worked because I just haven't had hands on.

 

I was doing lens equivalencies but, for anamorphic, I think it's more important to use the vertical size.  I wasn't really sure for the different options on the 5D what that measurement would be.

[nahua]

In the very early stages of development you could choose your aspect ratio, or more the vertical and horizontal resolution.  I believe you could get 2048 x 1920 too.  But they soon found out that the bitrate is what limits how much you can record.  In order to help people out, they changed it to presets with aspect ratios, then you can choose your resolution.  But they couldn't get consistent recording at 2048, so the horizontal is fixed at 1920.  Vertical was a problem too, so they limit it at 1440 I believe.  I think you can shoot 1920x1440 for a few seconds only.  The 3:2 sweetspot is 1920x1280 for continuous recording at 104MB/sec.  I haven't tried 1920x1440 because I can only get a few seconds.  I believe the bitrate is 125MB/sec way over the rates of any current CF cards.  And the buffer is too small to store so much information.  You can check out my video tests on Vimeo and download the video file if you want to take a look:

 

[nahua]

OK I'm looking at my ML settings now.  The vertical resolution is maxed out at 1288.  I don't know why, it must be in the forums somewhere.  Here are the resolutions:  1600x1200, 1728x1288, 1856x1288, 1920x1288.  However - there are resolutions from 2048 up to 3584, you just can't select them.  I believe the 4:3 setting I use 1920x1280 is near the max resolution with the most stable continuous recording.  If you want more answers maybe you can post in the ML forum or PM A1ex. I believe he's the main person responsible for the resolution settings.  I know you can modify the resolution if you change a file in the Raw_rec module, but you'd have to recompile it.  I don't know anything about that, maybe A1ex can do it.  I'd love to shoot 2048, at least get close to 2K.  But I don't think it's possible for continuous recording, again because of the recording bitrate would be too high.

[Sean Cunningham]

Thanks!

[dubzee]

The only way you can select those higher resolutions is by going into Crop Mode. To do so, make sure Magic Zoom is off and then press the Zoom button on the left of the camera. That takes a 1:1 crop from the centre of the sensor albeit at 3x zoom.

[Sean Cunningham]

Yeah, I wouldn't be at all interested in any 1:1 crop mode.

[dhessel]

With ML and card spanning turned on with a 5DIII people are getting about 110 MB/s so that is about the limit that camera can do continuously anyway. ML doesn't limit the maximum resolution in any way, those limits are a result of limits of the data stream that they hack into.

 

In non-crop mode the maximum resolution is the max resolution of the camera / 3 5760 x 3840 / 3 = 1920 x 1280. That is because the camera takes the raw sensor data and only reads out every third pixel and then sends that to compression. ML can grab that data after it has been down sampled and dump it to the card. They have no direct access to the sensor data so they are limited by what they can access which I believe is actually a raw data stream meant for the Live View.

 

In crop mode the data stream is much larger (3584 x 1440) and is every pixel instead of every third so that is the max resolution in that mode. Also the resolutions have limitations as well can cannot be just any random value. The width in pixels should be evenly divisible by 16 and the number if bits per horizontal line must be divisible by 16. If you do the math you will see that most of the possible resolutions are already included, I recently added one my self at 1792 wide.

 

So shooting roughly 4/3 aspect the highest non-crop mode res you can do is 1728 x 1280. In crop mode you could do 1920 x 1440.

[Sean Cunningham]

Okay, but resolution wasn't really my interest.  It's evident in all the testing that you get good enough resolution with a 5D, what's not clear to me is where this is being pulled from the sensor.  Since the sensor is 36mm x 24mm (1.5:1) it stands to reason that standard 16:9 motion video is derived by lopping off top and bottom.  So shooting standard 16:9 means you're pulling image from ~20.25mm sensor height.

 

What I'm really wanting to know, and perhaps I just didn't phrase my question right, when you define a non-16:9 aspect ratio, is the vertical dimension always pulled from the same ~20.25mm area as standard 16:9 imagery?  Between 1.78:1 and 1.5:1 it makes more logical sense to grow the crop region vertically until eventually pulling from the entire 24mm sensor height and  you aren't cropping any more.  Then between 1.5:1 and 1.33:1 (or 1.2:1 or 1:1) you maintain full 24mm sensor height and just keep cropping in the sides, using less and less of the 36mm sensor width.  But perhaps that isn't the way it's done for some technical reason and ~20.25mm is always read for the vertical dimension and all sub-16:9 aspect ratios are created by cropping from the 36mm dimension.

 

I'm just curious which way it works.

[dhessel]

If you are shooting in regular mode then ML always crops center. The resulting recording will be the width that you specify centered on the sensor matching the aspect ratio as close as possible by adjusting the height again centered. If you were to shoot with an aspect ration of 1 the resulting image size would be 1920 x 1280 since ML always matches width even if it cannot match the height.

 

If you were to shoot 960 x 640 you would be using a 18 mm x 12mm portion of the sensor centered on the sensor. This effect is easy to see in the live view when change resolutions.

 

In crop mode that may not be true, I know there was a time when it was cropping to the left but that may have changed. I personally never use crop mode.

[Sean Cunningham]

 

If you were to shoot 960 x 640 you would be using a 18 mm x 12mm portion of the sensor centered on the sensor. This effect is easy to see in the live view when change resolutions.

 

 

I think that's my answer.  And this would mean 960x1280 would be 18mm x 24mm, if you were so inclined.  That the height is variable under normal, non "crop mode" shooting is good to know.  Thanks!

[dhessel]

I think that's my answer.  And this would mean 960x1280 would be 18mm x 24mm, if you were so inclined.  That the height is variable under normal, non "crop mode" shooting is good to know.  Thanks!

 

Yes that is correct.

[nahua]

With ML and card spanning turned on with a 5DIII people are getting about 110 MB/s so that is about the limit that camera can do continuously anyway. ML doesn't limit the maximum resolution in any way, those limits are a result of limits of the data stream that they hack into.

 

In non-crop mode the maximum resolution is the max resolution of the camera / 3 5760 x 3840 / 3 = 1920 x 1280. That is because the camera takes the raw sensor data and only reads out every third pixel and then sends that to compression. ML can grab that data after it has been down sampled and dump it to the card. They have no direct access to the sensor data so they are limited by what they can access which I believe is actually a raw data stream meant for the Live View.

 

In crop mode the data stream is much larger (3584 x 1440) and is every pixel instead of every third so that is the max resolution in that mode. Also the resolutions have limitations as well can cannot be just any random value. The width in pixels should be evenly divisible by 16 and the number if bits per horizontal line must be divisible by 16. If you do the math you will see that most of the possible resolutions are already included, I recently added one my self at 1792 wide.

 

So shooting roughly 4/3 aspect the highest non-crop mode res you can do is 1728 x 1280. In crop mode you could do 1920 x 1440.

Thanks dhessel for clearing this up.  I really don't know the technical aspects.  I just shoot with it!  :-)